search

wangxiang1230 / OadTR

Code for our ICCV 2021 Paper "OadTR: Online Action Detection with Transformers".
MIT License
87 stars 12 forks source link

code question #19

Open 007invictus opened 2 years ago

007invictus commented 2 years ago

in 45and47 line of the document "dataset.py" did you want to write self.data_root instead pickle_root?

007invictus commented 2 years ago

and i met an error: Traceback (most recent call last): File "main.py", line 197, in main(args) File "main.py", line 150, in main args.clip_max_norm) File "/data/wtyuan/git/OadTR/train.py", line 65, in train_one_epoch model(camera_inputs, motion_inputs) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, *kwargs) File "/data/wtyuan/git/OadTR/transformer_models/ViT.py", line 160, in forward x = self.linear_encoding(x) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(input, **kwargs) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 103, in forward return F.linear(input, self.weight, self.bias) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/functional.py", line 1848, in linear return torch._C._nn.linear(input, weight, bias) RuntimeError: mat1 and mat2 shapes cannot be multiplied (8192x4096 and 3072x1024)

could you please give some advices?

wangxiang1230 commented 2 years ago

and i met an error: Traceback (most recent call last): File "main.py", line 197, in main(args) File "main.py", line 150, in main args.clip_max_norm) File "/data/wtyuan/git/OadTR/train.py", line 65, in train_one_epoch model(camera_inputs, motion_inputs) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(*input, *kwargs) File "/data/wtyuan/git/OadTR/transformer_models/ViT.py", line 160, in forward x = self.linear_encoding(x) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl return forward_call(input, **kwargs) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 103, in forward return F.linear(input, self.weight, self.bias) File "/data/wtyuan/anaconda3/lib/python3.7/site-packages/torch/nn/functional.py", line 1848, in linear return torch._C._nn.linear(input, weight, bias) RuntimeError: mat1 and mat2 shapes cannot be multiplied (8192x4096 and 3072x1024)

could you please give some advices?

It looks like you are using the Thumos14-Kinetics feature, which is 4096-dim (rgb: 2048, flow: 2048). You need to change the network input dimension by "python main.py --dim_feature 4096“.

wangxiang1230 commented 2 years ago

in 45and47 line of the document "dataset.py" did you want to write self.data_root instead pickle_root?

The purpose of these lines is to read the packed features, all packaged in a pickle file.