train YOLOX-S with RTMDet Trainer on COCO

README.md

@@ -158,7 +158,7 @@ python train.py --cuda -d coco --root path/to/COCO -m yolov1 -bs 16 --max_epoch
 | YOLOvx-N |  640  |  300  |                         |                    |                        |                   |                   |                    |  |
 | YOLOvx-S |  640  |  300  |                         |                    |                        |                   |                   |                    |  |
 | YOLOvx-M |  640  |  300  |                         |                    |                        |                   |                   |                    |  |
-| YOLOvx-L |  640  |  300  |         50.2            |       68.6         |          50.0          |        68.4       |      176.6        |        47.6        | [ckpt](https://github.com/yjh0410/PyTorch_YOLO_Tutorial/releases/download/yolo_tutorial_ckpt/yolovx_l_coco.pth) |
+| YOLOvx-L |  640  |  300  |         50.2            |        68.6        |          50.0          |        68.4       |      176.6        |        47.6        | [ckpt](https://github.com/yjh0410/PyTorch_YOLO_Tutorial/releases/download/yolo_tutorial_ckpt/yolovx_l_coco.pth) |
 
 * Redesigned RT-DETR:
 

train.sh

@@ -3,13 +3,13 @@ python train.py \
         --cuda \
         -d voc \
         --root /mnt/share/ssd2/dataset/ \
-        -m yolox_n \
-        -bs 16 \
+        -m yolox_s \
+        -bs 64 \
         -size 640 \
-        --wp_epoch 1 \
-        --max_epoch 5 \
-        --eval_epoch 1 \
-        --no_aug_epoch 2 \
+        --wp_epoch 3 \
+        --max_epoch 300 \
+        --eval_epoch 10 \
+        --no_aug_epoch 20 \
         --ema \
         --fp16 \
         --multi_scale \