GraphMLP Ground true dataset training and testing

Hello, thank you very much for your work. Enter 2D ground truth, execute: Python main.py -- frames 1-- batch_ Size 256-- keypoints' gt '. The best result obtained is p1: 37.22, which is not as good as 34.6 in the paper. May I ask what is causing this? Looking forward to your reply, thank you. @Vegetebird

Aug 12 '23 03:08 jianlai123-123

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

Nov 16 '24 13:11 zhang-yuxian

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

Why can't i get the weight file of the best model when i use "python main.py --frames 1 --batch_size 256" , after training for 20 epochs? Do i need to add something to save the best model in ''python main.py --frames 1 --batch_size 256''?

Nov 28 '24 07:11 grizzlies33

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

Why can't i get the weight file of the best model when i use "python main.py --frames 1 --batch_size 256" , after training for 20 epochs? Do i need to add something to save the best model in ''python main.py --frames 1 --batch_size 256''?

I have obtained the file generated below when I use "python main.py --frames 243 --batch_size 256" .

"python main.py --frames 243 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" .
Only keep the model with the smallest MPJPE value, delete others, and place the model generated in the pre training stage among them.
"python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 243".

Finally, can you provide relevant screenshots！

Nov 28 '24 08:11 zhang-yuxian

This is one of the files I obtained after using "python main.py --frames 243 --batch_size 256", and the train.log file is empty and this is my training process

Nov 28 '24 08:11 grizzlies33

I have obtained the file generated below when I use "python main.py --frames 243 --batch_size 256" .

"python main.py --frames 243 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" .
Only keep the model with the smallest MPJPE value, delete others, and place the model generated in the pre training stage among them.
"python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 243".

Finally, can you provide relevant screenshots！

At 2024-11-28 15:28:39, "grizzlies33" @.***> wrote:

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

Why can't i get the weight file of the best model when i use "python main.py --frames 1 --batch_size 256" , after training for 20 epochs? Do i need to add something to save the best model in ''python main.py --frames 1 --batch_size 256''?

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

Nov 28 '24 08:11 zhang-yuxian

This is one of the files I obtained after using "python main.py --frames 243 --batch_size 256", and the train.log file is empty and this is my training process

Your training file indicates 1 frame instead of 243 frames, and is this the Human3.6M dataset? My personal suggestion is to download the latest GraphMLP, then decompress and retrain it

Nov 28 '24 08:11 zhang-yuxian

yes, i used Human3.6M and ''python main.py --frames 1 --batch_size 256'' sorry i didn't look carefully. And I modified the GraphMLP model ,but i don't think that change will affect this. I will try again, thanks

Nov 28 '24 10:11 grizzlies33

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

How many epochs should i train by adding the refine module after pretraining for 20 epochs?

Dec 01 '24 04:12 grizzlies33

20

1, I suggest you find common/arguments.py; 2, "parser.add_argument('--keypoints', default='cpn_ft_h36m_dbb', type=str)"-----"parser.add_argument('--keypoints', default='gt', type=str)"; 3, "python main.py --frames 1 --batch_size 256" , After training for 20 epochs, add refine module , "python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [your best model saved path]" 4, "python main.py --test --previous_dir 'checkpoint/[test model file name]' --frames 1"

How many epochs should i train by adding the refine module after pretraining for 20 epochs?

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

Dec 01 '24 04:12 zhang-yuxian

hellozhang-yuxian,after train 20 epoch,and next ,I add refine model，and replace best saved model path to my path,but when I run, these are questions like follow: RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [256, 1536]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

Dec 30 '24 05:12 danwenxiu

When testing, the code will be run. It is necessary to ensure that the code during the run is consistent with the code at the time of testing.

hellozhang-yuxian,after train 20 epoch,and next ,I add refine model，and replace best saved model path to my path,but when I run, these are questions like follow: RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [256, 1536]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

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

Dec 30 '24 05:12 zhang-yuxian

hellozhang-yuxian,after train 20 epoch,and next ,I add refine model，and replace best saved model path to my path,but when I run, these are questions like follow: RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [256, 1536]], which is output 0 of ReluBackward0, is at version 1; expected version 0 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

I also encountered this problem, but I solved it.You only need to modify the eighteenth line "nn.Dropout(0.5,inplace=True)" in the refine.py file to "nn.Dropout(0.5,inplace=False)", and the problem can be solved. like this:

Apr 29 '25 12:04 zhangjinxiao134253

1、我建议你找common/arguments.py; 2， “parser.add_argument（'--keypoints'， default='cpn_ft_h36m_dbb'， type=str）”-----“parser.add_argument（'--keypoints'， default='gt'， type=str）”; 3、“python main.py --frames 1 --batch_size 256”，训练20个epoch后，添加refine模块，“python main.py --frames 1 --batch_size 256 --refine --lr 1e-5 --previous_dir [你最好的模型保存路径]”4、“python main.py --test --previous_dir 'checkpoint/[测试模型文件名]' --frames 1”

@zhang-yuxian Hello, The paper doesn't mention that the Ground truth 2D keypoints used the refinement module, right? It probably wasn't added.

Aug 22 '25 12:08 shahuang659

Hello, thank you very much for your work. Enter 2D ground truth, execute: Python main.py -- frames 1-- batch_ Size 256-- keypoints' gt '. The best result obtained is p1: 37.22, which is not as good as 34.6 in the paper. May I ask what is causing this? Looking forward to your reply, thank you. @Vegetebird

@jianlai123-123 Hello, have you solved it? Can 2DGT achieve 34.2 as mentioned in the paper?

Sep 01 '25 13:09 shahuang659