yihongxu / transcenter Goto Github PK

Do you use gradcam (I cannot visualize heatmap by this way) or have a special function

Hi, thanks for sharing the codes. Could you please release the previous version 1 ?

Thanks a lot !

Hello, thanks for your excellent work, and i would like to know the reason why the loss of tracking decoder is hard to decrease when applied on my custom small object dataset ?

there are pacakges installed from PyPI.
so when I use

conda create --name <env_name> --file requirements.txt

I get the following error:
PackagesNotFoundError: The following packages are not available from current channels:

Could you use the following commands to upload your environments?

conda env export > environment.yaml
conda env create -n <env_name> -f environment.yaml

No module named 'MultiScaleDeformableAttention'

I failed in this step:

I am getting the following error in the deformable convolution package.

error in modulated_deformable_im2col_cuda: no kernel image is available for execution on the device
error in modulated_deformable_im2col_cuda: no kernel image is available for execution on the device
error in modulated_deformable_im2col_cuda: no kernel image is available for execution on the device
Traceback (most recent call last):
File "/opt/conda/lib/python3.7/pdb.py", line 1699, in main
pdb._runscript(mainpyfile)
File "/opt/conda/lib/python3.7/pdb.py", line 1568, in _runscript
self.run(statement)
File "/opt/conda/lib/python3.7/bdb.py", line 578, in run
exec(cmd, globals, locals)
File "", line 1, in
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/main_mot17_tracking.py", line 32, in
import os
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/main_mot17_tracking.py", line 412, in main
model, criterion, data_loader_train, optimizer, device, epoch, args.clip_max_norm, adaptive_clip=args.adaptive_clip)
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/engine.py", line 65, in train_one_epoch
outputs = model(samples, pre_samples=pre_samples, pre_hm=pre_hm)
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/models/deformable_detr.py", line 287, in forward
hs[layer_lvl] = self.ida_up[0](hs[layer_lvl], 0, len(hs[layer_lvl]))[-1]
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/models/dla.py", line 98, in forward
layers[startp] = node(layers[startp])
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/net/coxfs01/srv/export/coxfs01/pfister_lab2/share_root/Lab/abanerjee/VideoProjects/TransCenter_official/training/transcenter/models/dla.py", line 48, in forward
x = self.conv(x)
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in call
result = self.forward(*input, **kwargs)
File "/opt/DCNv2/dcn_v2.py", line 128, in forward
self.deformable_groups)
File "/opt/DCNv2/dcn_v2.py", line 31, in forward
ctx.deformable_groups)
RuntimeError: cublas runtime error : the GPU program failed to execute at /opt/conda/conda-bld/pytorch_1591914880026/work/aten/src/THC/THCBlas.cu:335

I am using the pytorch1-5cuda10-1_RTX.sif image with V100 gpu. Is there a way to make it work?

Thanks

Hi,
I noticed that the config file (detracker_reidV3.yaml) contains a parameter about a motion model. I see that it is imported into the parameters of each new track (line 132 in track.py) but I don't see where this parameter is actually used, for example when predicting the new position of a track before the linear matching.
I also noticed that in the function tracks_dets_matching_tracking() function there is an update to the current track position but the line "t.prev_pos = t.pos" is commented out (line 184 in tracker.py).

Are you using the velocity/motion information in some other part of the code or did you decide not to use this information at all?

Traceback (most recent call last):
File "/home/valca509/Desktop/TransCenter_official-main/training/main_mot20.py", line 511, in
main(args)
File "/home/valca509/Desktop/TransCenter_official-main/training/main_mot20.py", line 366, in main
model, criterion, postprocessors, data_loader_val, base_ds, device, args.output_dir, args.half
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
return func(*args, **kwargs)
File "/home/valca509/Desktop/TransCenter_official-main/training/engine.py", line 172, in evaluate
outputs = model(samples, pre_samples=pre_samples, pre_hm=targets['pre_cts'].clone())
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/home/valca509/Desktop/TransCenter_official-main/training/deformable_detr.py", line 215, in forward
hs[layer_lvl] = self.ida_up(hs[layer_lvl], 0, len(hs[layer_lvl]))[-1]
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/home/valca509/Desktop/TransCenter_official-main/models/dla.py", line 96, in forward
layers[i-1] = node(layers[i] + layers[i - 1])
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/cuda/amp/autocast_mode.py", line 141, in decorate_autocast
return func(*args, **kwargs)
File "/home/valca509/Desktop/TransCenter_official-main/models/dla.py", line 47, in forward
x_out = self.conv(x)
File "/home/valca509/anaconda3/envs/transcenter3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/home/valca509/Desktop/TransCenter_official-main/to_install/DCNv2_1.9/dcn_v2.py", line 172, in forward
self.deformable_groups,
File "/home/valca509/Desktop/TransCenter_official-main/to_install/DCNv2_1.9/dcn_v2.py", line 39, in forward
ctx.deformable_groups,
RuntimeError: expected scalar type Float but found Half

Hey, Thanks for this great work.

Have you released any pretrained checkpoint for TransCenter?

Hi! Do you have any update on the code?

Hi,
Are the models trained on the full train set or only on 75% (while the last 25% remained for validation)?
I'm asking because the class "GenericDataset_val" has a default value of train_ratio=0.5 which leads to a 75%-25% partition in the evaluation, but it is not mentioned in the pretrained models section of the readme.

感谢开源，
作者有试过输入小一点分辨率的图片，或者其他减少显存占用的思路，实验效果如何吗
（）当前的模型在batch_size调到1依然需要超过12G显存）

I found the model is trained only in person category, can we train it with multi-class like person and car?

Is there any comparison about the TransCenter_V2 and TransCenter_Lite?

Hello, I followed the option2 to create an environment and converted MOT20 dataset to coco format.
Then I want to track MOT20 dataset
python tracking/transcenter/mot20_pub.py --data_dir=./MOT20
I got an error and did not know how to deal with it.

This doesn't seem to be the problem with my network

Dear author:
Thanks for sharing the great work. For a lot people are interested in your algorithm, the fast way to know it is just git clone and try the demo.py. Would you kindly share it here? thank you.

I'm interested in your work. but i only have a single rtx3090 machine.

Thanks for your wonderful work! Can someone tell me how targets['pre_cts'] was obtained?