fylwen / htt Goto Github PK

Hello, may I ask if the FPHA dataset of your HTT model is open-sourced for training

hello！for training，if i use 3090Ti，how many hours for an epoch

Hello, I want to use my camera to test.Do you have real-time demo? Thanks!

Hello, thanks for share !
I encountered an error while training:
Traceback (most recent call last):
File "/home/zhangjiabao/data/1/zhangjiabao/mesh/HTT/train.py", line 238, in
main(args)
File "/home/zhangjiabao/data/1/zhangjiabao/mesh/HTT/train.py", line 131, in main
epochpass.epoch_pass(
File "/home/zhangjiabao/data/1/zhangjiabao/mesh/HTT/netscripts/epochpass.py", line 50, in epoch_pass
loss, results, losses = model(batch)
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/parallel/data_parallel.py", line 168, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/parallel/data_parallel.py", line 178, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/parallel/parallel_apply.py", line 86, in parallel_apply
output.reraise()
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/_utils.py", line 434, in reraise
raise exception
RuntimeError: Caught RuntimeError in replica 0 on device 0.
Original Traceback (most recent call last):
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/parallel/parallel_apply.py", line 61, in _worker
output = module(*input, **kwargs)
File "/home/zhangjiabao/data/1/zhangjiabao/miniconda3/envs/handmesh/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "/home/zhangjiabao/data/1/zhangjiabao/mesh/HTT/models/htt.py", line 169, in forward
batch_seq_ain_feature=flatten_ain_feature.contiguous().view(-1,self.ntokens_action,flatten_ain_feature.shape[-1])
RuntimeError: shape '[-1, 128, 512]' is invalid for input of size 16384

The shape doesn't match. I think one of the parameters is self.ntokens_action, do I think the initial settings for this parameter are customized?
May I know how to solve the problem of parameter matching.

Thank you very much for your work！
I am currently facing some issues

how can I report the hand MEPE and action recall rate by referring to our submitted results？
how to generate the handposes.json file from eval.py

Hello, thanks for share !
I wonder if the model only support first-person view?
And is there any code for inferencing in wild video?

Hey, thank you for your work.

In utils.py->to25Dbranch, why inp_res = [256,256], but inp_res=[480, 270]? And, why the trans_factor =100, scale_factor = 0.0001?

Thank you, looking for your reply!

fixed it by replacing it with glob2

I think there's incompatible function in preprocess_utils.py 38 line.

• Original
  frame_path_dst = os.listdir(fhb_rgb_dst, subj, action, seq, "color", frame)

• After
  frame_path_dst = os.path.join(fhb_rgb_dst, subj, action, seq, "color", frame)

Hello, the link you have provided for the pre-trained model doesn't seem to be working. Could you please update it?