remove YOLOX2

config/__init__.py

@@ -85,7 +85,6 @@ from .model_config.yolov5_config import yolov5_cfg
 from .model_config.yolov7_config import yolov7_cfg
 from .model_config.yolov8_config import yolov8_cfg
 from .model_config.yolox_config import yolox_cfg
-from .model_config.yolox2_config import yolox2_cfg
 
 def build_model_config(args):
     print('==============================')
@@ -114,9 +113,6 @@ def build_model_config(args):
     # YOLOX
     elif args.model in ['yolox_n', 'yolox_s', 'yolox_m', 'yolox_l', 'yolox_x']:
         cfg = yolox_cfg[args.model]
-    # YOLOX2
-    elif args.model in ['yolox2_n', 'yolox2_s', 'yolox2_m', 'yolox2_l', 'yolox2_x']:
-        cfg = yolox2_cfg[args.model]
 
     return cfg
 

config/model_config/yolox2_config.py

@@ -1,287 +0,0 @@
-# YOLOX2 Config
-
-
-yolox2_cfg = {
-    'yolox2_n':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 0.25,
-        'depth': 0.34,
-        'ratio': 2.0,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_nano',
-        # ---------------- Assignment config ----------------
-        ## Matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-    'yolox2_t':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 0.375,
-        'depth': 0.34,
-        'ratio': 2.0,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_nano',
-        # ---------------- Assignment config ----------------
-        ## Matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-    'yolox2_s':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 0.50,
-        'depth': 0.34,
-        'ratio': 2.0,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_small',
-        # ---------------- Assignment config ----------------
-        ## matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-    'yolox2_m':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 0.75,
-        'depth': 0.67,
-        'ratio': 1.5,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_medium',
-        # ---------------- Assignment config ----------------
-        ## matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-    'yolox2_l':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 1.0,
-        'depth': 1.0,
-        'ratio': 1.0,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_large',
-        # ---------------- Assignment config ----------------
-        ## matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-    'yolox2_x':{
-        # ---------------- Model config ----------------
-        ## Backbone
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 1.25,
-        'depth': 1.34,
-        'ratio': 1.0,
-        'stride': [8, 16, 32],  # P3, P4, P5
-        'max_stride': 32,
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-        ## Neck: PaFPN
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        ## Head
-        'head': 'decoupled_head',
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_depthwise': False,
-        # ---------------- Train config ----------------
-        ## input
-        'multi_scale': [0.7, 1.25],   # 448 -> 800
-        'trans_type': 'yolox_huge',
-        # ---------------- Assignment config ----------------
-        ## matcher
-        'matcher': "aligned_simota",
-        'matcher_hpy': {'soft_center_radius': 3.0,
-                        'topk_candidates': 13},
-        # ---------------- Loss config ----------------
-        ## loss weight
-        'loss_cls_weight': 1.0,
-        'loss_box_weight': 2.0,
-        # ---------------- Train config ----------------
-        'trainer_type': 'rtcdet',
-    },
-
-}

models/detectors/__init__.py

@@ -11,7 +11,6 @@ from .yolov5.build import build_yolov5
 from .yolov7.build import build_yolov7
 from .yolov8.build import build_yolov8
 from .yolox.build import build_yolox
-from .yolox2.build import build_yolox2
 
 
 # build object detector
@@ -53,10 +52,6 @@ def build_model(args,
     elif args.model in ['yolox_n', 'yolox_s', 'yolox_m', 'yolox_l', 'yolox_x']:
         model, criterion = build_yolox(
             args, model_cfg, device, num_classes, trainable, deploy)
-    # YOLOX2
-    elif args.model in ['yolox2_n', 'yolox2_s', 'yolox2_m', 'yolox2_l', 'yolox2_x']:
-        model, criterion = build_yolox2(
-            args, model_cfg, device, num_classes, trainable, deploy)
 
     if trainable:
         # Load pretrained weight

models/detectors/yolox2/README.md

@@ -1,57 +0,0 @@
-# YOLOX2:
-
-|   Model  | Batch | Scale | AP<sup>val<br>0.5:0.95 | AP<sup>val<br>0.5 | FLOPs<br><sup>(G) | Params<br><sup>(M) | Weight |
-|----------|-------|-------|------------------------|-------------------|-------------------|--------------------|--------|
-| YOLOX2-N | 8xb16 |  640  |                        |                   |                   |                    |  |
-| YOLOX2-T | 8xb16 |  640  |                        |                   |                   |                    |  |
-| YOLOX2-S | 8xb16 |  640  |                        |                   |                   |                    |  |
-| YOLOX2-M | 8xb16 |  640  |                        |                   |                   |                    |  |
-| YOLOX2-L | 8xb16 |  640  |                        |                   |                   |                    |  |
-| YOLOX2-X | 8xb16 |  640  |                        |                   |                   |                    |  |
-<!-- | YOLOX2-S | 8xb16 |  640  |          42.0          |        60.2       |        27.6       |          9.2       | [ckpt](https://github.com/yjh0410/RT-ODLab/releases/download/yolo_tutorial_ckpt/yolox2_s_coco.pth) | -->
-
-- For training, we train YOLOX2 series with 300 epochs on COCO.
-- For data augmentation, we use the large scale jitter (LSJ), Mosaic augmentation and Mixup augmentation, following the YOLOX.
-- For optimizer, we use AdamW with weight decay 0.05 and base per image lr 0.001 / 64,.
-- For learning rate scheduler, we use Linear decay scheduler.
-
-## Train YOLOX2
-### Single GPU
-Taking training YOLOX2-S on COCO as the example,
-```Shell
-python train.py --cuda -d coco --root path/to/coco -m yolox2_s -bs 16 -size 640 --wp_epoch 3 --max_epoch 300 --eval_epoch 10 --no_aug_epoch 20 --ema --fp16 --multi_scale 
-```
-
-### Multi GPU
-Taking training YOLOX2-S on COCO as the example,
-```Shell
-python -m torch.distributed.run --nproc_per_node=8 train.py --cuda -dist -d coco --root /data/datasets/ -m yolox2_s -bs 128 -size 640 --wp_epoch 3 --max_epoch 300  --eval_epoch 10 --no_aug_epoch 20 --ema --fp16 --sybn --multi_scale --save_folder weights/ 
-```
-
-## Test YOLOX2
-Taking testing YOLOX2-S on COCO-val as the example,
-```Shell
-python test.py --cuda -d coco --root path/to/coco -m yolox2_s --weight path/to/yolox2_s.pth -size 640 -vt 0.4 --show 
-```
-
-## Evaluate YOLOX2
-Taking evaluating YOLOX2-S on COCO-val as the example,
-```Shell
-python eval.py --cuda -d coco-val --root path/to/coco -m yolox2_s --weight path/to/yolox2_s.pth 
-```
-
-## Demo
-### Detect with Image
-```Shell
-python demo.py --mode image --path_to_img path/to/image_dirs/ --cuda -m yolox2_s --weight path/to/weight -size 640 -vt 0.4 --show
-```
-
-### Detect with Video
-```Shell
-python demo.py --mode video --path_to_vid path/to/video --cuda -m yolox2_s --weight path/to/weight -size 640 -vt 0.4 --show --gif
-```
-
-### Detect with Camera
-```Shell
-python demo.py --mode camera --cuda -m yolox2_s --weight path/to/weight -size 640 -vt 0.4 --show --gif
-```

models/detectors/yolox2/build.py

@@ -1,44 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding:utf-8 -*-
-
-import torch
-import torch.nn as nn
-
-from .loss import build_criterion
-from .yolox2 import YOLOX2
-
-
-# build object detector
-def build_yolox2(args, cfg, device, num_classes=80, trainable=False, deploy=False):
-    print('==============================')
-    print('Build {} ...'.format(args.model.upper()))
-    
-    print('==============================')
-    print('Model Configuration: \n', cfg)
-    
-    # -------------- Build YOLO --------------
-    model = YOLOX2(cfg                = cfg,
-                   device             = device, 
-                   num_classes        = num_classes,
-                   trainable          = trainable,
-                   conf_thresh        = args.conf_thresh,
-                   nms_thresh         = args.nms_thresh,
-                   topk               = args.topk,
-                   deploy             = deploy,
-                   no_multi_labels    = args.no_multi_labels,
-                   nms_class_agnostic = args.nms_class_agnostic
-                   )
-
-    # -------------- Initialize YOLO --------------
-    for m in model.modules():
-        if isinstance(m, nn.BatchNorm2d):
-            m.eps = 1e-3
-            m.momentum = 0.03    
-            
-    # -------------- Build criterion --------------
-    criterion = None
-    if trainable:
-        # build criterion for training
-        criterion = build_criterion(args, cfg, device, num_classes)
-        
-    return model, criterion

models/detectors/yolox2/loss.py

@@ -1,184 +0,0 @@
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-from utils.box_ops import get_ious
-from utils.distributed_utils import get_world_size, is_dist_avail_and_initialized
-
-from .matcher import AlignedSimOTA
-
-
-class Criterion(object):
-    def __init__(self, args, cfg, device, num_classes=80):
-        self.args = args
-        self.cfg = cfg
-        self.device = device
-        self.num_classes = num_classes
-        self.max_epoch = args.max_epoch
-        self.no_aug_epoch = args.no_aug_epoch
-        self.aux_bbox_loss = False
-        # --------------- Loss config ---------------
-        self.loss_cls_weight = cfg['loss_cls_weight']
-        self.loss_box_weight = cfg['loss_box_weight']
-        # --------------- Matcher config ---------------
-        self.matcher_hpy = cfg['matcher_hpy']
-        self.matcher = AlignedSimOTA(soft_center_radius = self.matcher_hpy['soft_center_radius'],
-                                     topk_candidates    = self.matcher_hpy['topk_candidates'],
-                                     num_classes        = num_classes,
-                                     )
-
-    def loss_classes(self, pred_cls, target, beta=2.0):
-        # Quality FocalLoss
-        """
-            pred_cls: (torch.Tensor): [N, C]。
-            target:   (tuple([torch.Tensor], [torch.Tensor])): label -> (N,), score -> (N)
-        """
-        label, score = target
-        pred_sigmoid = pred_cls.sigmoid()
-        scale_factor = pred_sigmoid
-        zerolabel = scale_factor.new_zeros(pred_cls.shape)
-
-        ce_loss = F.binary_cross_entropy_with_logits(
-            pred_cls, zerolabel, reduction='none') * scale_factor.pow(beta)
-        
-        bg_class_ind = pred_cls.shape[-1]
-        pos = ((label >= 0) & (label < bg_class_ind)).nonzero().squeeze(1)
-        pos_label = label[pos].long()
-
-        scale_factor = score[pos] - pred_sigmoid[pos, pos_label]
-
-        ce_loss[pos, pos_label] = F.binary_cross_entropy_with_logits(
-            pred_cls[pos, pos_label], score[pos],
-            reduction='none') * scale_factor.abs().pow(beta)
-
-        return ce_loss
-    
-    def loss_bboxes(self, pred_box, gt_box):
-        ious = get_ious(pred_box, gt_box, box_mode="xyxy", iou_type='giou')
-        loss_box = 1.0 - ious
-
-        return loss_box
-    
-    def loss_bboxes_aux(self, pred_reg, gt_box, anchors, stride_tensors):
-        # xyxy -> cxcy&bwbh
-        gt_cxcy = (gt_box[..., :2] + gt_box[..., 2:]) * 0.5
-        gt_bwbh = gt_box[..., 2:] - gt_box[..., :2]
-        # encode gt box
-        gt_cxcy_encode = (gt_cxcy - anchors) / stride_tensors
-        gt_bwbh_encode = torch.log(gt_bwbh / stride_tensors)
-        gt_box_encode = torch.cat([gt_cxcy_encode, gt_bwbh_encode], dim=-1)
-        # l1 loss
-        loss_box_aux = F.l1_loss(pred_reg, gt_box_encode, reduction='none')
-
-        return loss_box_aux
-
-    def __call__(self, outputs, targets, epoch=0):        
-        """
-            outputs['pred_obj']: List(Tensor) [B, M, 1]
-            outputs['pred_cls']: List(Tensor) [B, M, C]
-            outputs['pred_box']: List(Tensor) [B, M, 4]
-            outputs['pred_box']: List(Tensor) [B, M, 4]
-            outputs['strides']: List(Int) [8, 16, 32] output stride
-            targets: (List) [dict{'boxes': [...], 
-                                 'labels': [...], 
-                                 'orig_size': ...}, ...]
-        """
-        bs = outputs['pred_cls'][0].shape[0]
-        device = outputs['pred_cls'][0].device
-        fpn_strides = outputs['strides']
-        anchors = outputs['anchors']
-        # preds: [B, M, C]
-        cls_preds = torch.cat(outputs['pred_cls'], dim=1)
-        box_preds = torch.cat(outputs['pred_box'], dim=1)
-        
-        # --------------- label assignment ---------------
-        cls_targets = []
-        box_targets = []
-        assign_metrics = []
-        for batch_idx in range(bs):
-            tgt_labels = targets[batch_idx]["labels"].to(device)  # [N,]
-            tgt_bboxes = targets[batch_idx]["boxes"].to(device)   # [N, 4]
-            assigned_result = self.matcher(fpn_strides=fpn_strides,
-                                           anchors=anchors,
-                                           pred_cls=cls_preds[batch_idx].detach(),
-                                           pred_box=box_preds[batch_idx].detach(),
-                                           gt_labels=tgt_labels,
-                                           gt_bboxes=tgt_bboxes
-                                           )
-            cls_targets.append(assigned_result['assigned_labels'])
-            box_targets.append(assigned_result['assigned_bboxes'])
-            assign_metrics.append(assigned_result['assign_metrics'])
-
-        # List[B, M, C] -> Tensor[BM, C]
-        cls_targets = torch.cat(cls_targets, dim=0)
-        box_targets = torch.cat(box_targets, dim=0)
-        assign_metrics = torch.cat(assign_metrics, dim=0)
-
-        # FG cat_id: [0, num_classes -1], BG cat_id: num_classes
-        bg_class_ind = self.num_classes
-        pos_inds = ((cls_targets >= 0) & (cls_targets < bg_class_ind)).nonzero().squeeze(1)
-        num_fgs = assign_metrics.sum()
-
-        if is_dist_avail_and_initialized():
-            torch.distributed.all_reduce(num_fgs)
-        num_fgs = (num_fgs / get_world_size()).clamp(1.0).item()
-
-        # ------------------ Classification loss ------------------
-        cls_preds = cls_preds.view(-1, self.num_classes)
-        loss_cls = self.loss_classes(cls_preds, (cls_targets, assign_metrics))
-        loss_cls = loss_cls.sum() / num_fgs
-
-        # ------------------ Regression loss ------------------
-        box_preds_pos = box_preds.view(-1, 4)[pos_inds]
-        box_targets_pos = box_targets[pos_inds]
-        loss_box = self.loss_bboxes(box_preds_pos, box_targets_pos)
-        loss_box = loss_box.sum() / num_fgs
-
-        # total loss
-        losses = self.loss_cls_weight * loss_cls + \
-                 self.loss_box_weight * loss_box
-
-        # ------------------ Aux regression loss ------------------
-        loss_box_aux = None
-        if epoch >= (self.max_epoch - self.no_aug_epoch - 1):
-            ## reg_preds
-            reg_preds = torch.cat(outputs['pred_reg'], dim=1)
-            reg_preds_pos = reg_preds.view(-1, 4)[pos_inds]
-            ## anchor tensors
-            anchors_tensors = torch.cat(outputs['anchors'], dim=0)[None].repeat(bs, 1, 1)
-            anchors_tensors_pos = anchors_tensors.view(-1, 2)[pos_inds]
-            ## stride tensors
-            stride_tensors = torch.cat(outputs['stride_tensors'], dim=0)[None].repeat(bs, 1, 1)
-            stride_tensors_pos = stride_tensors.view(-1, 1)[pos_inds]
-            ## aux loss
-            loss_box_aux = self.loss_bboxes_aux(reg_preds_pos, box_targets_pos, anchors_tensors_pos, stride_tensors_pos)
-            loss_box_aux = loss_box_aux.sum() / num_fgs
-
-            losses += loss_box_aux
-
-        # Loss dict
-        if loss_box_aux is None:
-            loss_dict = dict(
-                    loss_cls = loss_cls,
-                    loss_box = loss_box,
-                    losses = losses
-            )
-        else:
-            loss_dict = dict(
-                    loss_cls = loss_cls,
-                    loss_box = loss_box,
-                    loss_box_aux = loss_box_aux,
-                    losses = losses
-                    )
-
-        return loss_dict
-    
-
-def build_criterion(args, cfg, device, num_classes):
-    criterion = Criterion(args, cfg, device, num_classes)
-
-    return criterion
-
-
-if __name__ == "__main__":
-    pass

models/detectors/yolox2/matcher.py

@@ -1,162 +0,0 @@
-# ------------------------------------------------------------------------------------------
-# This code referenced to https://github.com/open-mmlab/mmyolo/models/task_modules/assigners/batch_dsl_assigner.py
-# ------------------------------------------------------------------------------------------
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-from utils.box_ops import box_iou
-
-
-# -------------------------- Aligned SimOTA assigner --------------------------
-class AlignedSimOTA(object):
-    def __init__(self, num_classes, soft_center_radius=3.0, topk_candidates=13):
-        self.num_classes = num_classes
-        self.soft_center_radius = soft_center_radius
-        self.topk_candidates = topk_candidates
-
-    @torch.no_grad()
-    def __call__(self, 
-                 fpn_strides, 
-                 anchors, 
-                 pred_cls, 
-                 pred_box, 
-                 gt_labels,
-                 gt_bboxes):
-        # [M,]
-        strides = torch.cat([torch.ones_like(anchor_i[:, 0]) * stride_i
-                                for stride_i, anchor_i in zip(fpn_strides, anchors)], dim=-1)
-        # List[F, M, 2] -> [M, 2]
-        num_gt = len(gt_labels)
-        anchors = torch.cat(anchors, dim=0)
-
-        # check gt
-        if num_gt == 0 or gt_bboxes.max().item() == 0.:
-            return {
-                'assigned_labels': gt_labels.new_full(pred_cls[..., 0].shape,
-                                                      self.num_classes,
-                                                      dtype=torch.long),
-                'assigned_bboxes': gt_bboxes.new_full(pred_box.shape, 0),
-                'assign_metrics': gt_bboxes.new_full(pred_cls[..., 0].shape, 0)
-            }
-        
-        # get inside points: [N, M]
-        is_in_gt = self.find_inside_points(gt_bboxes, anchors)
-        valid_mask = is_in_gt.sum(dim=0) > 0  # [M,]
-
-        # ----------------------------------- soft center prior -----------------------------------
-        gt_center = (gt_bboxes[..., :2] + gt_bboxes[..., 2:]) / 2.0
-        distance = (anchors.unsqueeze(0) - gt_center.unsqueeze(1)
-                    ).pow(2).sum(-1).sqrt() / strides.unsqueeze(0)  # [N, M]
-        distance = distance * valid_mask.unsqueeze(0)
-        soft_center_prior = torch.pow(10, distance - self.soft_center_radius)
-
-        # ----------------------------------- regression cost -----------------------------------
-        pair_wise_ious, _ = box_iou(gt_bboxes, pred_box)  # [N, M]
-        pair_wise_ious_loss = -torch.log(pair_wise_ious + 1e-8) * 3.0
-
-        # ----------------------------------- classification cost -----------------------------------
-        ## select the predicted scores corresponded to the gt_labels
-        pairwise_pred_scores = pred_cls.permute(1, 0)  # [M, C] -> [C, M]
-        pairwise_pred_scores = pairwise_pred_scores[gt_labels.long(), :].float()   # [N, M]
-        ## scale factor
-        scale_factor = (pair_wise_ious - pairwise_pred_scores.sigmoid()).abs().pow(2.0)
-        ## cls cost
-        pair_wise_cls_loss = F.binary_cross_entropy_with_logits(
-            pairwise_pred_scores, pair_wise_ious,
-            reduction="none") * scale_factor # [N, M]
-            
-        del pairwise_pred_scores
-
-        ## foreground cost matrix
-        cost_matrix = pair_wise_cls_loss + pair_wise_ious_loss + soft_center_prior
-        max_pad_value = torch.ones_like(cost_matrix) * 1e9
-        cost_matrix = torch.where(valid_mask[None].repeat(num_gt, 1),   # [N, M]
-                                  cost_matrix, max_pad_value)
-
-        # ----------------------------------- dynamic label assignment -----------------------------------
-        matched_pred_ious, matched_gt_inds, fg_mask_inboxes = self.dynamic_k_matching(
-            cost_matrix, pair_wise_ious, num_gt)
-        del pair_wise_cls_loss, cost_matrix, pair_wise_ious, pair_wise_ious_loss
-
-        # -----------------------------------process assigned labels -----------------------------------
-        assigned_labels = gt_labels.new_full(pred_cls[..., 0].shape,
-                                             self.num_classes)  # [M,]
-        assigned_labels[fg_mask_inboxes] = gt_labels[matched_gt_inds].squeeze(-1)
-        assigned_labels = assigned_labels.long()  # [M,]
-
-        assigned_bboxes = gt_bboxes.new_full(pred_box.shape, 0)        # [M, 4]
-        assigned_bboxes[fg_mask_inboxes] = gt_bboxes[matched_gt_inds]  # [M, 4]
-
-        assign_metrics = gt_bboxes.new_full(pred_cls[..., 0].shape, 0) # [M, 4]
-        assign_metrics[fg_mask_inboxes] = matched_pred_ious            # [M, 4]
-
-        assigned_dict = dict(
-            assigned_labels=assigned_labels,
-            assigned_bboxes=assigned_bboxes,
-            assign_metrics=assign_metrics
-            )
-        
-        return assigned_dict
-
-    def find_inside_points(self, gt_bboxes, anchors):
-        """
-            gt_bboxes: Tensor -> [N, 2]
-            anchors:   Tensor -> [M, 2]
-        """
-        num_anchors = anchors.shape[0]
-        num_gt = gt_bboxes.shape[0]
-
-        anchors_expand = anchors.unsqueeze(0).repeat(num_gt, 1, 1)           # [N, M, 2]
-        gt_bboxes_expand = gt_bboxes.unsqueeze(1).repeat(1, num_anchors, 1)  # [N, M, 4]
-
-        # offset
-        lt = anchors_expand - gt_bboxes_expand[..., :2]
-        rb = gt_bboxes_expand[..., 2:] - anchors_expand
-        bbox_deltas = torch.cat([lt, rb], dim=-1)
-
-        is_in_gts = bbox_deltas.min(dim=-1).values > 0
-
-        return is_in_gts
-    
-    def dynamic_k_matching(self, cost_matrix, pairwise_ious, num_gt):
-        """Use IoU and matching cost to calculate the dynamic top-k positive
-        targets.
-
-        Args:
-            cost_matrix (Tensor): Cost matrix.
-            pairwise_ious (Tensor): Pairwise iou matrix.
-            num_gt (int): Number of gt.
-            valid_mask (Tensor): Mask for valid bboxes.
-        Returns:
-            tuple: matched ious and gt indexes.
-        """
-        matching_matrix = torch.zeros_like(cost_matrix, dtype=torch.uint8)
-        # select candidate topk ious for dynamic-k calculation
-        candidate_topk = min(self.topk_candidates, pairwise_ious.size(1))
-        topk_ious, _ = torch.topk(pairwise_ious, candidate_topk, dim=1)
-        # calculate dynamic k for each gt
-        dynamic_ks = torch.clamp(topk_ious.sum(1).int(), min=1)
-
-        # sorting the batch cost matirx is faster than topk
-        _, sorted_indices = torch.sort(cost_matrix, dim=1)
-        for gt_idx in range(num_gt):
-            topk_ids = sorted_indices[gt_idx, :dynamic_ks[gt_idx]]
-            matching_matrix[gt_idx, :][topk_ids] = 1
-
-        del topk_ious, dynamic_ks, topk_ids
-
-        prior_match_gt_mask = matching_matrix.sum(0) > 1
-        if prior_match_gt_mask.sum() > 0:
-            cost_min, cost_argmin = torch.min(
-                cost_matrix[:, prior_match_gt_mask], dim=0)
-            matching_matrix[:, prior_match_gt_mask] *= 0
-            matching_matrix[cost_argmin, prior_match_gt_mask] = 1
-
-        # get foreground mask inside box and center prior
-        fg_mask_inboxes = matching_matrix.sum(0) > 0
-        matched_pred_ious = (matching_matrix *
-                             pairwise_ious).sum(0)[fg_mask_inboxes]
-        matched_gt_inds = matching_matrix[:, fg_mask_inboxes].argmax(0)
-
-        return matched_pred_ious, matched_gt_inds, fg_mask_inboxes
-    

models/detectors/yolox2/yolox2.py

@@ -1,203 +0,0 @@
-# --------------- Torch components ---------------
-import torch
-import torch.nn as nn
-
-# --------------- Model components ---------------
-from .yolox2_backbone import build_backbone
-from .yolox2_neck import build_neck
-from .yolox2_pafpn import build_fpn
-from .yolox2_head import build_det_head
-from .yolox2_pred import build_pred_layer
-
-# --------------- External components ---------------
-from utils.misc import multiclass_nms
-
-
-# YOLOX2
-class YOLOX2(nn.Module):
-    def __init__(self,
-                 cfg,
-                 device,
-                 num_classes = 20,
-                 conf_thresh = 0.01,
-                 nms_thresh  = 0.5,
-                 topk        = 1000,
-                 trainable   = False,
-                 deploy      = False,
-                 no_multi_labels = False,
-                 nms_class_agnostic = False):
-        super(YOLOX2, self).__init__()
-        # ---------------------- Basic Parameters ----------------------
-        self.cfg = cfg
-        self.device = device
-        self.strides = cfg['stride']
-        self.num_classes = num_classes
-        self.trainable = trainable
-        self.conf_thresh = conf_thresh
-        self.nms_thresh = nms_thresh
-        self.num_levels = len(self.strides)
-        self.num_classes = num_classes
-        self.topk_candidates = topk
-        self.deploy = deploy
-        self.no_multi_labels = no_multi_labels
-        self.nms_class_agnostic = nms_class_agnostic
-        self.head_dim = round(256 * cfg['width'])
-        
-        # ---------------------- Network Parameters ----------------------
-        ## ----------- Backbone -----------
-        self.backbone, feat_dims = build_backbone(cfg)
-
-        ## ----------- Neck: SPP -----------
-        self.neck = build_neck(cfg, feat_dims[-1], feat_dims[-1])
-        feat_dims[-1] = self.neck.out_dim
-        
-        ## ----------- Neck: FPN -----------
-        self.fpn = build_fpn(cfg, feat_dims, out_dim=self.head_dim)
-        self.fpn_dims = self.fpn.out_dim
-
-        ## ----------- Heads -----------
-        self.det_heads = build_det_head(cfg, self.fpn_dims, self.head_dim, self.num_levels)
-
-        ## ----------- Preds -----------
-        self.pred_layers = build_pred_layer(cls_dim     = self.det_heads.cls_head_dim,
-                                            reg_dim     = self.det_heads.reg_head_dim,
-                                            strides     = self.strides,
-                                            num_classes = num_classes,
-                                            num_coords  = 4,
-                                            num_levels  = self.num_levels)
-
-    ## post-process
-    def post_process(self, cls_preds, box_preds):
-        """
-        Input:
-            cls_preds: List[np.array] -> [[M, C], ...]
-            box_preds: List[np.array] -> [[M, 4], ...]
-        Output:
-            bboxes: np.array -> [N, 4]
-            scores: np.array -> [N,]
-            labels: np.array -> [N,]
-        """
-        assert len(cls_preds) == self.num_levels
-        all_scores = []
-        all_labels = []
-        all_bboxes = []
-        
-        for cls_pred_i, box_pred_i in zip(cls_preds, box_preds):
-            cls_pred_i = cls_pred_i[0]
-            box_pred_i = box_pred_i[0]
-            if self.no_multi_labels:
-                # [M,]
-                scores, labels = torch.max(cls_pred_i.sigmoid(), dim=1)
-
-                # Keep top k top scoring indices only.
-                num_topk = min(self.topk_candidates, box_pred_i.size(0))
-
-                # topk candidates
-                predicted_prob, topk_idxs = scores.sort(descending=True)
-                topk_scores = predicted_prob[:num_topk]
-                topk_idxs = topk_idxs[:num_topk]
-
-                # filter out the proposals with low confidence score
-                keep_idxs = topk_scores > self.conf_thresh
-                scores = topk_scores[keep_idxs]
-                topk_idxs = topk_idxs[keep_idxs]
-
-                labels = labels[topk_idxs]
-                bboxes = box_pred_i[topk_idxs]
-            else:
-                # [M, C] -> [MC,]
-                scores_i = cls_pred_i.sigmoid().flatten()
-
-                # Keep top k top scoring indices only.
-                num_topk = min(self.topk_candidates, box_pred_i.size(0))
-
-                # torch.sort is actually faster than .topk (at least on GPUs)
-                predicted_prob, topk_idxs = scores_i.sort(descending=True)
-                topk_scores = predicted_prob[:num_topk]
-                topk_idxs = topk_idxs[:num_topk]
-
-                # filter out the proposals with low confidence score
-                keep_idxs = topk_scores > self.conf_thresh
-                scores = topk_scores[keep_idxs]
-                topk_idxs = topk_idxs[keep_idxs]
-
-                anchor_idxs = torch.div(topk_idxs, self.num_classes, rounding_mode='floor')
-                labels = topk_idxs % self.num_classes
-
-                bboxes = box_pred_i[anchor_idxs]
-
-            all_scores.append(scores)
-            all_labels.append(labels)
-            all_bboxes.append(bboxes)
-
-        scores = torch.cat(all_scores, dim=0)
-        labels = torch.cat(all_labels, dim=0)
-        bboxes = torch.cat(all_bboxes, dim=0)
-
-        # to cpu & numpy
-        scores = scores.cpu().numpy()
-        labels = labels.cpu().numpy()
-        bboxes = bboxes.cpu().numpy()
-
-        # nms
-        scores, labels, bboxes = multiclass_nms(
-            scores, labels, bboxes, self.nms_thresh, self.num_classes, self.nms_class_agnostic)
-
-        return bboxes, scores, labels
-    
-    # ---------------------- Main Process for Inference ----------------------
-    @torch.no_grad()
-    def inference_single_image(self, x):
-        # ---------------- Backbone ----------------
-        pyramid_feats = self.backbone(x)
-
-        # ---------------- Neck: SPP ----------------
-        pyramid_feats[-1] = self.neck(pyramid_feats[-1])
-
-        # ---------------- Neck: PaFPN ----------------
-        pyramid_feats = self.fpn(pyramid_feats)
-
-        # ---------------- Heads ----------------
-        cls_feats, reg_feats = self.det_heads(pyramid_feats)
-
-        # ---------------- Preds ----------------
-        outputs = self.pred_layers(cls_feats, reg_feats)
-
-        all_cls_preds = outputs['pred_cls']
-        all_box_preds = outputs['pred_box']
-
-        if self.deploy:
-            cls_preds = torch.cat(all_cls_preds, dim=1)[0]
-            box_preds = torch.cat(all_box_preds, dim=1)[0]
-            scores = cls_preds.sigmoid()
-            bboxes = box_preds
-            # [n_anchors_all, 4 + C]
-            outputs = torch.cat([bboxes, scores], dim=-1)
-
-            return outputs
-        else:
-            # post process
-            bboxes, scores, labels = self.post_process(all_cls_preds, all_box_preds)
-        
-            return bboxes, scores, labels
-
-    def forward(self, x):
-        if not self.trainable:
-            return self.inference_single_image(x)
-        else:
-            # ---------------- Backbone ----------------
-            pyramid_feats = self.backbone(x)
-
-            # ---------------- Neck: SPP ----------------
-            pyramid_feats[-1] = self.neck(pyramid_feats[-1])
-
-            # ---------------- Neck: PaFPN ----------------
-            pyramid_feats = self.fpn(pyramid_feats)
-
-            # ---------------- Heads ----------------
-            cls_feats, reg_feats = self.det_heads(pyramid_feats)
-
-            # ---------------- Preds ----------------
-            outputs = self.pred_layers(cls_feats, reg_feats)
-            
-            return outputs 

models/detectors/yolox2/yolox2_backbone.py

@@ -1,116 +0,0 @@
-import torch
-import torch.nn as nn
-
-try:
-    from .yolox2_basic import Conv, Yolov8StageBlock
-except:
-    from yolox2_basic import Conv, Yolov8StageBlock
-
-
-# ---------------------------- Basic functions ----------------------------
-## ELAN-CSPNet
-class Yolov8Backbone(nn.Module):
-    def __init__(self, width=1.0, depth=1.0, ratio=1.0, act_type='silu', norm_type='BN', depthwise=False):
-        super(Yolov8Backbone, self).__init__()
-        self.feat_dims = [round(64 * width), round(128 * width), round(256 * width), round(512 * width), round(512 * width * ratio)]
-        # P1/2
-        self.layer_1 = Conv(3, self.feat_dims[0], k=3, p=1, s=2, act_type=act_type, norm_type=norm_type)
-        # P2/4
-        self.layer_2 = nn.Sequential(
-            Conv(self.feat_dims[0], self.feat_dims[1], k=3, p=1, s=2, act_type=act_type, norm_type=norm_type),
-            Yolov8StageBlock(in_dim     = self.feat_dims[1],
-                             out_dim    = self.feat_dims[1],
-                             num_blocks = round(3*depth),
-                             shortcut   = True,
-                             act_type   = act_type,
-                             norm_type  = norm_type,
-                             depthwise  = depthwise)
-        )
-        # P3/8
-        self.layer_3 = nn.Sequential(
-            Conv(self.feat_dims[1], self.feat_dims[2], k=3, p=1, s=2, act_type=act_type, norm_type=norm_type),
-            Yolov8StageBlock(in_dim     = self.feat_dims[2],
-                             out_dim    = self.feat_dims[2],
-                             num_blocks = round(6*depth),
-                             shortcut   = True,
-                             act_type   = act_type,
-                             norm_type  = norm_type,
-                             depthwise  = depthwise)
-        )
-        # P4/16
-        self.layer_4 = nn.Sequential(
-            Conv(self.feat_dims[2], self.feat_dims[3], k=3, p=1, s=2, act_type=act_type, norm_type=norm_type),
-            Yolov8StageBlock(in_dim     = self.feat_dims[3],
-                             out_dim    = self.feat_dims[3],
-                             num_blocks = round(6*depth),
-                             shortcut   = True,
-                             act_type   = act_type,
-                             norm_type  = norm_type,
-                             depthwise  = depthwise)
-        )
-        # P5/32
-        self.layer_5 = nn.Sequential(
-            Conv(self.feat_dims[3], self.feat_dims[4], k=3, p=1, s=2, act_type=act_type, norm_type=norm_type),
-            Yolov8StageBlock(in_dim     = self.feat_dims[4],
-                             out_dim    = self.feat_dims[4],
-                             num_blocks = round(3*depth),
-                             shortcut   = True,
-                             act_type   = act_type,
-                             norm_type  = norm_type,
-                             depthwise  = depthwise)
-        )
-
-    def forward(self, x):
-        c1 = self.layer_1(x)
-        c2 = self.layer_2(c1)
-        c3 = self.layer_3(c2)
-        c4 = self.layer_4(c3)
-        c5 = self.layer_5(c4)
-
-        outputs = [c3, c4, c5]
-
-        return outputs
-
-
-# ---------------------------- Functions ----------------------------
-## build Yolov8's Backbone
-def build_backbone(cfg): 
-    # model
-    backbone = Yolov8Backbone(width=cfg['width'],
-                              depth=cfg['depth'],
-                              ratio=cfg['ratio'],
-                              act_type=cfg['bk_act'],
-                              norm_type=cfg['bk_norm'],
-                              depthwise=cfg['bk_depthwise']
-                              )
-    feat_dims = backbone.feat_dims[-3:]
-        
-    return backbone, feat_dims
-
-
-if __name__ == '__main__':
-    import time
-    from thop import profile
-    cfg = {
-        'bk_act': 'silu',
-        'bk_norm': 'BN',
-        'bk_depthwise': False,
-        'width': 1.0,
-        'depth': 1.0,
-        'ratio': 1.0,
-    }
-    model, feats = build_backbone(cfg)
-    x = torch.randn(1, 3, 640, 640)
-    t0 = time.time()
-    outputs = model(x)
-    t1 = time.time()
-    print('Time: ', t1 - t0)
-    for out in outputs:
-        print(out.shape)
-
-    x = torch.randn(1, 3, 640, 640)
-    print('==============================')
-    flops, params = profile(model, inputs=(x, ), verbose=False)
-    print('==============================')
-    print('GFLOPs : {:.2f}'.format(flops / 1e9 * 2))
-    print('Params : {:.2f} M'.format(params / 1e6))

models/detectors/yolox2/yolox2_basic.py

@@ -1,137 +0,0 @@
-import torch
-import torch.nn as nn
-
-
-# --------------------- Basic modules ---------------------
-class SiLU(nn.Module):
-    """export-friendly version of nn.SiLU()"""
-
-    @staticmethod
-    def forward(x):
-        return x * torch.sigmoid(x)
-
-def get_conv2d(c1, c2, k, p, s, d, g, bias=False):
-    conv = nn.Conv2d(c1, c2, k, stride=s, padding=p, dilation=d, groups=g, bias=bias)
-
-    return conv
-
-def get_activation(act_type=None):
-    if act_type == 'relu':
-        return nn.ReLU(inplace=True)
-    elif act_type == 'lrelu':
-        return nn.LeakyReLU(0.1, inplace=True)
-    elif act_type == 'mish':
-        return nn.Mish(inplace=True)
-    elif act_type == 'silu':
-        return nn.SiLU(inplace=True)
-    elif act_type is None:
-        return nn.Identity()
-    else:
-        raise NotImplementedError
-        
-def get_norm(norm_type, dim):
-    if norm_type == 'BN':
-        return nn.BatchNorm2d(dim)
-    elif norm_type == 'GN':
-        return nn.GroupNorm(num_groups=32, num_channels=dim)
-    elif norm_type is None:
-        return nn.Identity()
-    else:
-        raise NotImplementedError
-
-class Conv(nn.Module):
-    def __init__(self, 
-                 c1,                   # in channels
-                 c2,                   # out channels 
-                 k=1,                  # kernel size 
-                 p=0,                  # padding
-                 s=1,                  # padding
-                 d=1,                  # dilation
-                 act_type='lrelu',     # activation
-                 norm_type='BN',       # normalization
-                 depthwise=False):
-        super(Conv, self).__init__()
-        convs = []
-        add_bias = False if norm_type else True
-        if depthwise:
-            convs.append(get_conv2d(c1, c1, k=k, p=p, s=s, d=d, g=c1, bias=add_bias))
-            # depthwise conv
-            if norm_type:
-                convs.append(get_norm(norm_type, c1))
-            if act_type:
-                convs.append(get_activation(act_type))
-            # pointwise conv
-            convs.append(get_conv2d(c1, c2, k=1, p=0, s=1, d=d, g=1, bias=add_bias))
-            if norm_type:
-                convs.append(get_norm(norm_type, c2))
-            if act_type:
-                convs.append(get_activation(act_type))
-
-        else:
-            convs.append(get_conv2d(c1, c2, k=k, p=p, s=s, d=d, g=1, bias=add_bias))
-            if norm_type:
-                convs.append(get_norm(norm_type, c2))
-            if act_type:
-                convs.append(get_activation(act_type))
-            
-        self.convs = nn.Sequential(*convs)
-
-
-    def forward(self, x):
-        return self.convs(x)
-
-
-# --------------------- Yolov8 modules ---------------------
-## Yolov8 BottleNeck
-class Yolov8Bottleneck(nn.Module):
-    def __init__(self,
-                 in_dim,
-                 out_dim,
-                 expand_ratio = 0.5,
-                 kernel_sizes = [3, 3],
-                 shortcut     = True,
-                 act_type     = 'silu',
-                 norm_type    = 'BN',
-                 depthwise    = False,):
-        super(Yolov8Bottleneck, self).__init__()
-        inter_dim = int(out_dim * expand_ratio)  # hidden channels            
-        self.cv1 = Conv(in_dim, inter_dim, k=kernel_sizes[0], p=kernel_sizes[0]//2, norm_type=norm_type, act_type=act_type, depthwise=depthwise)
-        self.cv2 = Conv(inter_dim, out_dim, k=kernel_sizes[1], p=kernel_sizes[1]//2, norm_type=norm_type, act_type=act_type, depthwise=depthwise)
-        self.shortcut = shortcut and in_dim == out_dim
-
-    def forward(self, x):
-        h = self.cv2(self.cv1(x))
-
-        return x + h if self.shortcut else h
-
-# Yolov8 StageBlock
-class Yolov8StageBlock(nn.Module):
-    def __init__(self,
-                 in_dim,
-                 out_dim,
-                 num_blocks = 1,
-                 shortcut   = False,
-                 act_type   = 'silu',
-                 norm_type  = 'BN',
-                 depthwise  = False,):
-        super(Yolov8StageBlock, self).__init__()
-        self.inter_dim = out_dim // 2
-        self.input_proj = Conv(in_dim, out_dim, k=1, act_type=act_type, norm_type=norm_type)
-        self.m = nn.Sequential(*(
-            Yolov8Bottleneck(self.inter_dim, self.inter_dim, 1.0, [3, 3], shortcut, act_type, norm_type, depthwise)
-            for _ in range(num_blocks)))
-        self.output_proj = Conv((2 + num_blocks) * self.inter_dim, out_dim, k=1, act_type=act_type, norm_type=norm_type)
-
-    def forward(self, x):
-        # Input proj
-        x1, x2 = torch.chunk(self.input_proj(x), 2, dim=1)
-        out = list([x1, x2])
-
-        # Bottlenecl
-        out.extend(m(out[-1]) for m in self.m)
-
-        # Output proj
-        out = self.output_proj(torch.cat(out, dim=1))
-
-        return out
-    

models/detectors/yolox2/yolox2_head.py

@@ -1,164 +0,0 @@
-import torch
-import torch.nn as nn
-
-try:
-    from .yolox2_basic import Conv
-except:
-    from yolox2_basic import Conv
-
-
-# Single-level Head
-class SingleLevelHead(nn.Module):
-    def __init__(self,
-                 in_dim       :int  = 256,
-                 cls_head_dim :int  = 256,
-                 reg_head_dim :int  = 256,
-                 num_cls_head :int  = 2,
-                 num_reg_head :int  = 2,
-                 act_type     :str  = "silu",
-                 norm_type    :str  = "BN",
-                 depthwise    :bool = False):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.in_dim = in_dim
-        self.num_cls_head = num_cls_head
-        self.num_reg_head = num_reg_head
-        self.act_type = act_type
-        self.norm_type = norm_type
-        self.depthwise = depthwise
-        
-        # --------- Network Parameters ----------
-        ## cls head
-        cls_feats = []
-        self.cls_head_dim = cls_head_dim
-        for i in range(num_cls_head):
-            if i == 0:
-                cls_feats.append(
-                    Conv(in_dim, self.cls_head_dim, k=3, p=1, s=1, 
-                         act_type=act_type,
-                         norm_type=norm_type,
-                         depthwise=depthwise)
-                        )
-            else:
-                cls_feats.append(
-                    Conv(self.cls_head_dim, self.cls_head_dim, k=3, p=1, s=1, 
-                        act_type=act_type,
-                        norm_type=norm_type,
-                        depthwise=depthwise)
-                        )      
-        ## reg head
-        reg_feats = []
-        self.reg_head_dim = reg_head_dim
-        for i in range(num_reg_head):
-            if i == 0:
-                reg_feats.append(
-                    Conv(in_dim, self.reg_head_dim, k=3, p=1, s=1, 
-                         act_type=act_type,
-                         norm_type=norm_type,
-                         depthwise=depthwise)
-                        )
-            else:
-                reg_feats.append(
-                    Conv(self.reg_head_dim, self.reg_head_dim, k=3, p=1, s=1, 
-                         act_type=act_type,
-                         norm_type=norm_type,
-                         depthwise=depthwise)
-                        )
-        self.cls_feats = nn.Sequential(*cls_feats)
-        self.reg_feats = nn.Sequential(*reg_feats)
-
-        self.init_weights()
-        
-    def init_weights(self):
-        """Initialize the parameters."""
-        for m in self.modules():
-            if isinstance(m, torch.nn.Conv2d):
-                # In order to be consistent with the source code,
-                # reset the Conv2d initialization parameters
-                m.reset_parameters()
-
-    def forward(self, x):
-        """
-            in_feats: (Tensor) [B, C, H, W]
-        """
-        cls_feats = self.cls_feats(x)
-        reg_feats = self.reg_feats(x)
-
-        return cls_feats, reg_feats
-    
-# Multi-level Head
-class MultiLevelHead(nn.Module):
-    def __init__(self, cfg, in_dims, out_dim, num_levels=3):
-        super().__init__()
-        ## ----------- Network Parameters -----------
-        self.multi_level_heads = nn.ModuleList(
-            [SingleLevelHead(in_dim=in_dims[level],
-                             cls_head_dim = out_dim,
-                             reg_head_dim = out_dim,
-                             num_cls_head = cfg['num_cls_head'],
-                             num_reg_head = cfg['num_reg_head'],
-                             act_type     = cfg['head_act'],
-                             norm_type    = cfg['head_norm'],
-                             depthwise    = cfg['head_depthwise'])
-                             for level in range(num_levels)
-                             ])
-        # --------- Basic Parameters ----------
-        self.in_dims = in_dims
-        self.cls_head_dim = self.multi_level_heads[0].cls_head_dim
-        self.reg_head_dim = self.multi_level_heads[0].reg_head_dim
-
-
-    def forward(self, feats):
-        """
-            feats: List[(Tensor)] [[B, C, H, W], ...]
-        """
-        cls_feats = []
-        reg_feats = []
-        for feat, head in zip(feats, self.multi_level_heads):
-            # ---------------- Pred ----------------
-            cls_feat, reg_feat = head(feat)
-
-            cls_feats.append(cls_feat)
-            reg_feats.append(reg_feat)
-
-        return cls_feats, reg_feats
-    
-
-# build detection head
-def build_det_head(cfg, in_dims, out_dim, num_levels=3):
-    if cfg['head'] == 'decoupled_head':
-        head = MultiLevelHead(cfg, in_dims, out_dim, num_levels)
-
-    return head
-
-
-if __name__ == '__main__':
-    import time
-    from thop import profile
-    cfg = {
-        'head': 'decoupled_head',
-        'num_cls_head': 2,
-        'num_reg_head': 2,
-        'head_act': 'silu',
-        'head_norm': 'BN',
-        'head_depthwise': False,
-        'reg_max': 16,
-    }
-    fpn_dims = [256, 256, 256]
-    out_dim = 256
-    # Head-1
-    model = build_det_head(cfg, fpn_dims, out_dim, num_levels=3)
-    print(model)
-    fpn_feats = [torch.randn(1, fpn_dims[0], 80, 80), torch.randn(1, fpn_dims[1], 40, 40), torch.randn(1, fpn_dims[2], 20, 20)]
-    t0 = time.time()
-    outputs = model(fpn_feats)
-    t1 = time.time()
-    print('Time: ', t1 - t0)
-    # for out in outputs:
-    #     print(out.shape)
-
-    print('==============================')
-    flops, params = profile(model, inputs=(fpn_feats, ), verbose=False)
-    print('==============================')
-    print('Head-1: GFLOPs : {:.2f}'.format(flops / 1e9 * 2))
-    print('Head-1: Params : {:.2f} M'.format(params / 1e6))

models/detectors/yolox2/yolox2_neck.py

@@ -1,104 +0,0 @@
-import torch
-import torch.nn as nn
-
-try:
-    from .yolox2_basic import Conv
-except:
-    from yolox2_basic import Conv
-
-
-# Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
-class SPPF(nn.Module):
-    """
-        This code referenced to https://github.com/ultralytics/yolov5
-    """
-    def __init__(self, cfg, in_dim, out_dim, expand_ratio=0.5):
-        super().__init__()
-        inter_dim = int(in_dim * expand_ratio)
-        self.out_dim = out_dim
-        self.cv1 = Conv(in_dim, inter_dim, k=1, act_type=cfg['neck_act'], norm_type=cfg['neck_norm'])
-        self.cv2 = Conv(inter_dim * 4, out_dim, k=1, act_type=cfg['neck_act'], norm_type=cfg['neck_norm'])
-        self.m = nn.MaxPool2d(kernel_size=cfg['pooling_size'], stride=1, padding=cfg['pooling_size'] // 2)
-
-    def forward(self, x):
-        x = self.cv1(x)
-        y1 = self.m(x)
-        y2 = self.m(y1)
-
-        return self.cv2(torch.cat((x, y1, y2, self.m(y2)), 1))
-
-
-# SPPF block with CSP module
-class SPPFBlockCSP(nn.Module):
-    """
-        CSP Spatial Pyramid Pooling Block
-    """
-    def __init__(self, cfg, in_dim, out_dim, expand_ratio):
-        super(SPPFBlockCSP, self).__init__()
-        inter_dim = int(in_dim * expand_ratio)
-        self.out_dim = out_dim
-        self.cv1 = Conv(in_dim, inter_dim, k=1, act_type=cfg['neck_act'], norm_type=cfg['neck_norm'])
-        self.cv2 = Conv(in_dim, inter_dim, k=1, act_type=cfg['neck_act'], norm_type=cfg['neck_norm'])
-        self.m = nn.Sequential(
-            Conv(inter_dim, inter_dim, k=3, p=1, 
-                 act_type=cfg['neck_act'], norm_type=cfg['neck_norm'], 
-                 depthwise=cfg['neck_depthwise']),
-            SPPF(cfg, inter_dim, inter_dim, expand_ratio=1.0),
-            Conv(inter_dim, inter_dim, k=3, p=1, 
-                 act_type=cfg['neck_act'], norm_type=cfg['neck_norm'], 
-                 depthwise=cfg['neck_depthwise'])
-        )
-        self.cv3 = Conv(inter_dim * 2, self.out_dim, k=1, act_type=cfg['neck_act'], norm_type=cfg['neck_norm'])
-
-        
-    def forward(self, x):
-        x1 = self.cv1(x)
-        x2 = self.cv2(x)
-        x3 = self.m(x2)
-        y = self.cv3(torch.cat([x1, x3], dim=1))
-
-        return y
-
-
-def build_neck(cfg, in_dim, out_dim):
-    model = cfg['neck']
-    print('==============================')
-    print('Neck: {}'.format(model))
-    # build neck
-    if model == 'sppf':
-        neck = SPPF(cfg, in_dim, out_dim, cfg['neck_expand_ratio'])
-    elif model == 'csp_sppf':
-        neck = SPPFBlockCSP(cfg, in_dim, out_dim, cfg['neck_expand_ratio'])
-
-    return neck
-
-
-if __name__ == '__main__':
-    import time
-    from thop import profile
-    cfg = {
-        ## Neck: SPP
-        'neck': 'sppf',
-        'neck_expand_ratio': 0.5,
-        'pooling_size': 5,
-        'neck_act': 'silu',
-        'neck_norm': 'BN',
-        'neck_depthwise': False,
-    }
-    in_dim = 512
-    out_dim = 512
-    # Head-1
-    model = build_neck(cfg, in_dim, out_dim)
-    feat = torch.randn(1, in_dim, 20, 20)
-    t0 = time.time()
-    outputs = model(feat)
-    t1 = time.time()
-    print('Time: ', t1 - t0)
-    # for out in outputs:
-    #     print(out.shape)
-
-    print('==============================')
-    flops, params = profile(model, inputs=(feat, ), verbose=False)
-    print('==============================')
-    print('FPN: GFLOPs : {:.2f}'.format(flops / 1e9 * 2))
-    print('FPN: Params : {:.2f} M'.format(params / 1e6))

models/detectors/yolox2/yolox2_pafpn.py

@@ -1,169 +0,0 @@
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-try:
-    from .yolox2_basic import Conv, Yolov8StageBlock
-except:
-    from yolox2_basic import Conv, Yolov8StageBlock
-
-
-# PaFPN-ELAN
-class Yolox2PaFPN(nn.Module):
-    def __init__(self, 
-                 in_dims   = [256, 512, 512],
-                 out_dim   = None,
-                 width     = 1.0,
-                 depth     = 1.0,
-                 ratio     = 1.0,
-                 act_type  = 'silu',
-                 norm_type = 'BN',
-                 depthwise = False):
-        super(Yolox2PaFPN, self).__init__()
-        print('==============================')
-        print('FPN: {}'.format("Yolov8 PaFPN"))
-        # ---------------- Basic parameters ----------------
-        self.in_dims = in_dims
-        self.width = width
-        self.depth = depth
-        c3, c4, c5 = in_dims
-
-        # ---------------- Top dwon ----------------
-        ## P5 -> P4
-        self.top_down_layer_1 = Yolov8StageBlock(in_dim       = c5 + c4,
-                                                 out_dim      = round(512*width),
-                                                 num_blocks   = round(3*depth),
-                                                 shortcut     = False,
-                                                 act_type     = act_type,
-                                                 norm_type    = norm_type,
-                                                 depthwise    = depthwise,
-                                                 )
-        ## P4 -> P3
-        self.top_down_layer_2 = Yolov8StageBlock(in_dim       = round(512*width) + c3,
-                                                 out_dim      = round(256*width),
-                                                 num_blocks   = round(3*depth),
-                                                 shortcut     = False,
-                                                 act_type     = act_type,
-                                                 norm_type    = norm_type,
-                                                 depthwise    = depthwise,
-                                                 )
-        # ---------------- Bottom up ----------------
-        ## P3 -> P4
-        self.dowmsample_layer_1 = Conv(round(256*width), round(256*width), k=3, p=1, s=2, act_type=act_type, norm_type=norm_type, depthwise=depthwise)
-        self.bottom_up_layer_1 = Yolov8StageBlock(in_dim       = round(256*width) + round(512*width),
-                                                  out_dim      = round(512*width),
-                                                  num_blocks   = round(3*depth),
-                                                  shortcut     = False,
-                                                  act_type     = act_type,
-                                                  norm_type    = norm_type,
-                                                  depthwise    = depthwise,
-                                                  )
-        ## P4 -> P5
-        self.dowmsample_layer_2 = Conv(round(512*width), round(512*width), k=3, p=1, s=2, act_type=act_type, norm_type=norm_type, depthwise=depthwise)
-        self.bottom_up_layer_2 = Yolov8StageBlock(in_dim       = round(512 * width) + c5,
-                                                  out_dim      = round(512 * width * ratio),
-                                                  num_blocks   = round(3*depth),
-                                                  shortcut     = False,
-                                                  act_type     = act_type,
-                                                  norm_type    = norm_type,
-                                                  depthwise    = depthwise,
-                                                  )
-        ## output proj layers
-        if out_dim is not None:
-            self.out_layers = nn.ModuleList([
-                Conv(in_dim, out_dim, k=1, act_type=act_type, norm_type=norm_type)
-                     for in_dim in [round(256*width), round(512*width), round(512 * width * ratio)]
-                     ])
-            self.out_dim = [out_dim] * 3
-        else:
-            self.out_layers = None
-            self.out_dim = [round(256*width), round(512*width), round(512 * width * ratio)]
-
-        self.init_weights()
-        
-    def init_weights(self):
-        """Initialize the parameters."""
-        for m in self.modules():
-            if isinstance(m, torch.nn.Conv2d):
-                # In order to be consistent with the source code,
-                # reset the Conv2d initialization parameters
-                m.reset_parameters()
-
-    def forward(self, features):
-        c3, c4, c5 = features
-
-        # Top down
-        ## P5 -> P4
-        c6 = F.interpolate(c5, scale_factor=2.0)
-        c7 = torch.cat([c6, c4], dim=1)
-        c8 = self.top_down_layer_1(c7)
-        ## P4 -> P3
-        c9 = F.interpolate(c8, scale_factor=2.0)
-        c10 = torch.cat([c9, c3], dim=1)
-        c11 = self.top_down_layer_2(c10)
-
-        # Bottom up
-        # p3 -> P4
-        c12 = self.dowmsample_layer_1(c11)
-        c13 = torch.cat([c12, c8], dim=1)
-        c14 = self.bottom_up_layer_1(c13)
-        # P4 -> P5
-        c15 = self.dowmsample_layer_2(c14)
-        c16 = torch.cat([c15, c5], dim=1)
-        c17 = self.bottom_up_layer_2(c16)
-
-        out_feats = [c11, c14, c17] # [P3, P4, P5]
-        
-        # output proj layers
-        if self.out_layers is not None:
-            out_feats_proj = []
-            for feat, layer in zip(out_feats, self.out_layers):
-                out_feats_proj.append(layer(feat))
-            return out_feats_proj
-
-        return out_feats
-
-
-def build_fpn(cfg, in_dims, out_dim=None):
-    model = cfg['fpn']
-    # build neck
-    if model == 'yolox2_pafpn':
-        fpn_net = Yolox2PaFPN(in_dims   = in_dims,
-                              out_dim   = out_dim,
-                              width     = cfg['width'],
-                              depth     = cfg['depth'],
-                              ratio     = cfg['ratio'],
-                              act_type  = cfg['fpn_act'],
-                              norm_type = cfg['fpn_norm'],
-                              depthwise = cfg['fpn_depthwise']
-                              )
-    return fpn_net
-
-
-if __name__ == '__main__':
-    import time
-    from thop import profile
-    cfg = {
-        'fpn': 'yolox2_pafpn',
-        'fpn_act': 'silu',
-        'fpn_norm': 'BN',
-        'fpn_depthwise': False,
-        'width': 1.0,
-        'depth': 1.0,
-        'ratio': 1.0
-    }
-    fpn_dims = [256, 512, 512]
-    out_dim=256
-    model = build_fpn(cfg, fpn_dims, out_dim)
-    pyramid_feats = [torch.randn(1, fpn_dims[0], 80, 80), torch.randn(1, fpn_dims[1], 40, 40), torch.randn(1, fpn_dims[2], 20, 20)]
-    t0 = time.time()
-    outputs = model(pyramid_feats)
-    t1 = time.time()
-    print('Time: ', t1 - t0)
-    for out in outputs:
-        print(out.shape)
-
-    print('==============================')
-    flops, params = profile(model, inputs=(pyramid_feats, ), verbose=False)
-    print('==============================')
-    print('GFLOPs : {:.2f}'.format(flops / 1e9 * 2))
-    print('Params : {:.2f} M'.format(params / 1e6))

models/detectors/yolox2/yolox2_pred.py

@@ -1,154 +0,0 @@
-import math
-import torch
-import torch.nn as nn
-
-
-# Single-level pred layer
-class SingleLevelPredLayer(nn.Module):
-    def __init__(self,
-                 cls_dim     :int = 256,
-                 reg_dim     :int = 256,
-                 stride      :int = 32,
-                 num_classes :int = 80,
-                 num_coords  :int = 4):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.stride = stride
-        self.cls_dim = cls_dim
-        self.reg_dim = reg_dim
-        self.num_classes = num_classes
-        self.num_coords = num_coords
-
-        # --------- Network Parameters ----------
-        self.cls_pred = nn.Conv2d(cls_dim, num_classes, kernel_size=1)
-        self.reg_pred = nn.Conv2d(reg_dim, num_coords, kernel_size=1)                
-
-        self.init_bias()
-        
-    def init_bias(self):
-        # cls pred bias
-        b = self.cls_pred.bias.view(1, -1)
-        b.data.fill_(math.log(5 / self.num_classes / (640. / self.stride) ** 2))
-        self.cls_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
-        # reg pred bias
-        b = self.reg_pred.bias.view(-1, )
-        b.data.fill_(1.0)
-        self.reg_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
-
-    def generate_anchors(self, fmp_size):
-        """
-            fmp_size: (List) [H, W]
-        """
-        # generate grid cells
-        fmp_h, fmp_w = fmp_size
-        anchor_y, anchor_x = torch.meshgrid([torch.arange(fmp_h), torch.arange(fmp_w)])
-        # [H, W, 2] -> [HW, 2]
-        anchors = torch.stack([anchor_x, anchor_y], dim=-1).float().view(-1, 2)
-        anchors += 0.5  # add center offset
-        anchors *= self.stride
-
-        return anchors
-        
-    def forward(self, cls_feat, reg_feat):
-        # pred
-        cls_pred = self.cls_pred(cls_feat)
-        reg_pred = self.reg_pred(reg_feat)
-
-        # generate anchor boxes: [M, 4]
-        B, _, H, W = cls_pred.size()
-        fmp_size = [H, W]
-        anchors = self.generate_anchors(fmp_size)
-        anchors = anchors.to(cls_pred.device)
-        # stride tensor: [M, 1]
-        stride_tensor = torch.ones_like(anchors[..., :1]) * self.stride
-        
-        # [B, C, H, W] -> [B, H, W, C] -> [B, M, C]
-        cls_pred = cls_pred.permute(0, 2, 3, 1).contiguous().view(B, -1, self.num_classes)
-        reg_pred = reg_pred.permute(0, 2, 3, 1).contiguous().view(B, -1, 4)
-
-        # ---------------- Decode bbox ----------------
-        ctr_pred = reg_pred[..., :2] * self.stride + anchors[..., :2]
-        wh_pred = torch.exp(reg_pred[..., 2:]) * self.stride
-        pred_x1y1 = ctr_pred - wh_pred * 0.5
-        pred_x2y2 = ctr_pred + wh_pred * 0.5
-        box_pred = torch.cat([pred_x1y1, pred_x2y2], dim=-1)
-
-        # output dict
-        outputs = {"pred_cls": cls_pred,             # (Tensor) [B, M, C]
-                   "pred_reg": reg_pred,             # (Tensor) [B, M, 4]
-                   "pred_box": box_pred,             # (Tensor) [B, M, 4] 
-                   "anchors": anchors,               # (Tensor) [M, 2]
-                   "stride": self.stride,            # (Int)
-                   "stride_tensors": stride_tensor   # List(Tensor) [M, 1]
-                   }
-
-        return outputs
-
-# Multi-level pred layer
-class MultiLevelPredLayer(nn.Module):
-    def __init__(self,
-                 cls_dim,
-                 reg_dim,
-                 strides,
-                 num_classes :int = 80,
-                 num_coords  :int = 4,
-                 num_levels  :int = 3):
-        super().__init__()
-        # --------- Basic Parameters ----------
-        self.cls_dim = cls_dim
-        self.reg_dim = reg_dim
-        self.strides = strides
-        self.num_classes = num_classes
-        self.num_coords = num_coords
-        self.num_levels = num_levels
-
-        # ----------- Network Parameters -----------
-        ## multi-level pred layers
-        self.multi_level_preds = nn.ModuleList(
-            [SingleLevelPredLayer(cls_dim     = cls_dim,
-                                  reg_dim     = reg_dim,
-                                  stride      = strides[level],
-                                  num_classes = num_classes,
-                                  num_coords  = num_coords)
-                                  for level in range(num_levels)
-                                  ])
-        
-    def forward(self, cls_feats, reg_feats):
-        all_anchors = []
-        all_strides = []
-        all_cls_preds = []
-        all_box_preds = []
-        all_reg_preds = []
-        for level in range(self.num_levels):
-            # ---------------- Single level prediction ----------------
-            outputs = self.multi_level_preds[level](cls_feats[level], reg_feats[level])
-
-            # collect results
-            all_cls_preds.append(outputs["pred_cls"])
-            all_box_preds.append(outputs["pred_box"])
-            all_reg_preds.append(outputs["pred_reg"])
-            all_anchors.append(outputs["anchors"])
-            all_strides.append(outputs["stride_tensors"])
-        
-        # output dict
-        outputs = {"pred_cls": all_cls_preds,      # List(Tensor) [B, M, C]
-                   "pred_box": all_box_preds,      # List(Tensor) [B, M, 4]
-                   "pred_reg": all_reg_preds,      # List(Tensor) [B, M, 4]
-                   "anchors": all_anchors,         # List(Tensor) [M, 2]
-                   "strides": self.strides,        # List(Int) [8, 16, 32]
-                   "stride_tensors": all_strides   # List(Tensor) [M, 1]
-                   }
-
-        return outputs
-    
-
-# build detection head
-def build_pred_layer(cls_dim, reg_dim, strides, num_classes, num_coords=4, num_levels=3):
-    pred_layers = MultiLevelPredLayer(cls_dim     = cls_dim,
-                                      reg_dim     = reg_dim,
-                                      strides     = strides,
-                                      num_classes = num_classes,
-                                      num_coords  = num_coords,
-                                      num_levels  = num_levels) 
-
-    return pred_layers