YuHengsss / YOLOV

This repo is an implementation of PyTorch version YOLOV Series
Apache License 2.0
323 stars 46 forks source link

Can you tell me how to solve this problem? I use a new model and pre trained weights yolov++to train my dataset #81

Open andynnnnn opened 5 months ago

andynnnnn commented 5 months ago

Traceback (most recent call last):

File "D:/Github/YOLOV-master/YOLOV-master/tools\train.py", line 140, in args=(exp, args), │ └ Namespace(batch_size=16, cache=False, ckpt='D:\Github\YOLOV-master\YOLOV-master\weights\V++_FocalL.pth', devices=1, dist... └ ╒════════════════════╤═══════════════════════════════════════════════════════════════════════════════════════════════════════...

File "D:\Github\YOLOV-master\YOLOV-master\yolox\core\launch.py", line 98, in launch main_func(*args) │ └ (╒════════════════════╤══════════════════════════════════════════════════════════════════════════════════════════════════════... └ <function main at 0x0000014A201048B8>

File "D:/Github/YOLOV-master/YOLOV-master/tools\train.py", line 117, in main trainer.train() │ └ <function Trainer.train at 0x0000014A27B1C798> └ <yolox.core.trainer.Trainer object at 0x0000014A28AB4F88>

File "D:\Github\YOLOV-master\YOLOV-master\yolox\core\trainer.py", line 74, in train self.before_train() │ └ <function Trainer.before_train at 0x0000014A28AAE798> └ <yolox.core.trainer.Trainer object at 0x0000014A28AB4F88>

File "D:\Github\YOLOV-master\YOLOV-master\yolox\core\trainer.py", line 175, in before_train batch_size=self.args.batch_size, is_distributed=self.is_distributed │ │ │ │ └ False │ │ │ └ <yolox.core.trainer.Trainer object at 0x0000014A28AB4F88> │ │ └ 16 │ └ Namespace(batch_size=16, cache=False, ckpt='D:\Github\YOLOV-master\YOLOV-master\weights\V++_FocalL.pth', devices=1, dist... └ <yolox.core.trainer.Trainer object at 0x0000014A28AB4F88>

TypeError: get_evaluator() got an unexpected keyword argument 'batch_size'

YuHengsss commented 5 months ago

If you want to train the model on your own dataset, here are steps:

  1. Train the yolox base detector. In your case, you should modify the https://github.com/YuHengsss/YOLOV/blob/master/exps/swin_base/focal_large_base_vid.py and init the prertain weight using YOLOX_FocalLarge. However, this weight is not released yet. You may try YOLOX_SwinTiny with corresponding exp. Use the train.py in tools in this step to train the base detector.

  2. Train the V++ detector with exps in https://github.com/YuHengsss/YOLOV/tree/master/exps/yolov%2B%2B and use the pretrain weight produced by step 1. Use the vid_train.py in tools in this step to train the video detector.

andynnnnn commented 5 months ago

Thank you for your explanation; you've really helped me a lot!

andynnnnn commented 5 months ago

Hello again, I hope I'm not being too much of a bother. I've successfully run models through YOLOXS_VID and YOLOV_S on my own dataset, but I've noticed that the mAP50 on YOLOXS_VID is significantly higher than YOLOV_S (approximately 4 points higher). What should I do to improve the performance of YOLOV_S? Or should I consider trying a model with a larger parameter count?

YuHengsss commented 5 months ago

May I take a look at your training log on yolox and yolov please?

---- Replied Message ---- From @.> Date 06/06/2024 16:21 To @.> Cc @.>@.> Subject Re: [YuHengsss/YOLOV] Can you tell me how to solve this problem? I use a new model and pre trained weights yolov++to train my dataset (Issue #81)

Reopened #81. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

multi-type commented 1 month ago

@YuHengsss How did you obtain YOLOX-Swin Pretain Models on ImageNet VID? Please teach me.