RT-ODLab/models/detectors/yolov2/build.py

#!/usr/bin/env python3
# -*- coding:utf-8 -*-

import torch
import torch.nn as nn

from .loss import build_criterion
from .yolov2 import YOLOv2


# build object detector
def build_yolov2(args, cfg, device, num_classes=80, trainable=False):
    print('==============================')
    print('Build {} ...'.format(args.model.upper()))
    
    print('==============================')
    print('Model Configuration: \n', cfg)
    
    # -------------- Build YOLO --------------
    model = YOLOv2(
        cfg=cfg,
        device=device, 
        num_classes=num_classes,
        trainable=trainable,
        conf_thresh=args.conf_thresh,
        nms_thresh=args.nms_thresh,
        topk=args.topk,
        )

    # -------------- Initialize YOLO --------------
    # Init bias
    init_prob = 0.01
    bias_value = -torch.log(torch.tensor((1. - init_prob) / init_prob))
    # obj pred
    b = model.obj_pred.bias.view(1, -1)
    b.data.fill_(bias_value.item())
    model.obj_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
    # cls pred
    b = model.cls_pred.bias.view(1, -1)
    b.data.fill_(bias_value.item())
    model.cls_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
    # reg pred
    b = model.reg_pred.bias.view(-1, )
    b.data.fill_(1.0)
    model.reg_pred.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
    w = model.reg_pred.weight
    w.data.fill_(0.)
    model.reg_pred.weight = torch.nn.Parameter(w, requires_grad=True)


    # -------------- Build criterion --------------
    criterion = None
    if trainable:
        # build criterion for training
        criterion = build_criterion(cfg, device, num_classes)
    return model, criterion