can't creat training dataset

[ycwei0321]

Hi

I'm going to creat the training dataset. However, it gave me such error. see details below.

Running on Windows
C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\ltr\external-win does not exist. No changes made.
Warning: Cannot import panoptic_inference module, will use fallback method
Traceback (most recent call last):
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\create_training_dataset.py", line 305, in create_training_dataset
run_tracking(run_tracking_params)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\run_tracker.py", line 125, in run_tracking
early_stop_flag = run_tracker(args.tracker_name, args.tracker_param, args.runid, args.dataset_name, seq_name, args.debug,
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\run_tracker.py", line 75, in run_tracker
early_stop_flag = run_dataset(sequence_obj, trackers, debug, threads, visdom_info=visdom_info,search_scale = search_scale,target_sz_bias = target_sz_bias,obj_id=obj_id,gui_param = gui_param)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\evaluation\running.py", line 156, in run_dataset
early_stop_flag = run_sequence(seq, tracker_info, debug=debug, visdom_info=visdom_info,search_scale = search_scale,target_sz_bias = target_sz_bias, object_id=obj_id,gui_param = gui_param)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\evaluation\running.py", line 108, in run_sequence
output, early_stop_flag = tracker.run_sequence(seq, debug=debug, visdom_info=visdom_info,search_scale = search_scale,target_sz_bias = target_sz_bias,gui_param = gui_param)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\evaluation\tracker_multi.py", line 279, in run_sequence
output, early_stop_flag = self._track_sequence(tracker, seq, init_info, tracker_compensate = tracker_compensate,gui_param = gui_param)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking\evaluation\tracker_multi.py", line 379, in _track_sequence
out = tracker[i].initialize(image, init_info)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking..\pytracking\tracker\trdimp\trdimp.py", line 129, in initialize
self.init_classifier(init_backbone_feat)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking..\pytracking\tracker\trdimp\trdimp.py", line 1115, in init_classifier
self.target_filter, _, losses = self.net.classifier.get_filter(x, target_boxes, num_iter=num_iter,
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking..\ltr\models\target_classifier\linear_filter.py", line 115, in get_filter
weights = self.filter_initializer(feat, bb)
File "C:\Users\shahlab\anaconda3\envs\udmt\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking..\ltr\models\target_classifier\initializer.py", line 166, in forward
weights = self.filter_pool(feat, bb)
File "C:\Users\shahlab\anaconda3\envs\udmt\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\pytracking..\ltr\models\target_classifier\initializer.py", line 47, in forward
return self.prroi_pool(feat, roi1)
File "C:\Users\shahlab\anaconda3\envs\udmt\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\ltr\external\PreciseRoIPooling\pytorch\prroi_pool\prroi_pool.py", line 28, in forward
return prroi_pool2d(features, rois, self.pooled_height, self.pooled_width, self.spatial_scale)
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\ltr\external\PreciseRoIPooling\pytorch\prroi_pool\functional.py", line 48, in forward
_prroi_pooling = _import_prroi_pooling()
File "C:\Users\shahlab\UDMT\udmt\gui\tabs\ST_Net\ltr\external\PreciseRoIPooling\pytorch\prroi_pool\functional.py", line 26, in _import_prroi_pooling
_prroi_pooling = importlib.util.module_from_spec(spec)
File "", line 556, in module_from_spec
File "", line 1166, in create_module
File "", line 219, in _call_with_frames_removed
ImportError: DLL load failed while importing prroi_pool: The specified procedure could not be found.

I checked the folder, and it did not have external-win, see below:

How can I resolve it?

Thanks.

[ycwei0321]

I also tried to run the analyze vidoes with your pretrained model, but it's not working. see errors below:

Is this error caused by the lack of external-win? Thanks.

Best
Yichao

[cabooster]

Hi, thanks for the detailed report.

This issue is different from the model-download/checkpoint-loading problem. From your traceback, the error is caused by the Precise ROI Pooling dependency used by ST_Net / pytracking. In particular, the final error is:

ImportError: DLL load failed while importing prroi_pool: The specified procedure could not be found.

This usually means that the prroi_pool / PreciseRoIPooling extension was not correctly compiled or installed for your current Windows + PyTorch + CUDA environment. The message about external-win also suggests that the Windows binary dependency is missing or not properly configured.

Installing Precise ROI Pooling on Windows can be difficult because it depends on the exact versions of PyTorch, CUDA, and the C++/CUDA compiler toolchain. If you have access to a Linux server, we strongly recommend using the Linux version of UDMT instead. The environment setup is generally easier on Linux, and it also allows UDMT to run with GPU acceleration more reliably.

So the recommended solution is:

1. Set up UDMT on a Linux server following the Linux installation instructions.
2. Make sure CUDA and PyTorch are correctly installed.
3. Run the UDMT GUI / tracking pipeline from the Linux environment.

If you still need to use Windows, you would need to manually compile and install the PreciseRoIPooling extension for your exact environment, but this is more error-prone and we do not recommend it as the first option.