after running inference, instances for ALL videos showing on ONE video

[sophiaanner]

I have a set of videos on which I am trying to predict instances based on a previous model I had configured. This is a workaround I am trying because I am not able to open the project that I was previously working on in the new version of SLEAP (#2417).

I'm running sleap v1.5.1 with python 3.11 and a NVIDIA RTX PRO 500 blackwell GPU.

When I run inference, no matter what I select for "Predict on...", the predicted instances get placed on the first video in my list. If I select "20 frames in current video", it works as expected. But if I navigate to a new video (e.g. the 2nd on the list) and input the same, it adds instances for 20 frames to the first video on my list BUT with the predictions for the fly body parts of video 2. The same goes if I try to predict on just one frame.

I've attached a picture where I predicted on 340 frames (20 frames x 17 videos) and it placed ALL predictions onto the first video. As you can see, the instances do not line up with the flies in the video shown.

Here is a selection of the text in the command prompt. Any ideas how to resolve this?

--- Logging error in Loguru Handler #1 ---
Record was: {'elapsed': datetime.timedelta(seconds=4, microseconds=964509), 'exception': None, 'extra': {}, 'file': (name='legacy_models.py', path='C:\Users\sophi\AppData\Roaming\uv\tools\sleap\Lib\site-packages\sleap_nn\legacy_models.py'), 'function': 'load_legacy_model_weights', 'level': (name='INFO', no=20, icon='ℹ️'), 'line': 439, 'message': '✓ All weight assignments verified successfully', 'module': 'legacy_models', 'name': 'sleap_nn.legacy_models', 'process': (id=11932, name='MainProcess'), 'thread': (id=19004, name='MainThread'), 'time': datetime(2025, 10, 23, 11, 45, 22, 763103, tzinfo=datetime.timezone(datetime.timedelta(days=-1, seconds=61200), 'Pacific Daylight Time'))}
Traceback (most recent call last):
File "C:\Users\sophi\AppData\Roaming\uv\tools\sleap\Lib\site-packages\loguru_handler.py", line 206, in emit
self._sink.write(str_record)
File "C:\Users\sophi\AppData\Roaming\uv\tools\sleap\Lib\site-packages\loguru_simple_sinks.py", line 123, in write
self.function(message)
File "C:\Users\sophi\AppData\Roaming\uv\tools\sleap\Lib\site-packages\sleap_nn_init.py", line 29, in
lambda msg: print(msg, end=""),
^^^^^^^^^^^^^^^^^^
File "C:\Users\sophi\AppData\Roaming\uv\python\cpython-3.11.14-windows-x86_64-none\Lib\encodings\cp1252.py", line 19, in encode
return codecs.charmap_encode(input,self.errors,encoding_table)[0]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
UnicodeEncodeError: 'charmap' codec can't encode character '\u2713' in position 84: character maps to
--- End of logging error ---
2025-10-23 11:45:22 | INFO | sleap_nn.predict:run_inference:319 | Started inference at: 2025-10-23 11:45:22.068304
2025-10-23 11:45:22 | INFO | sleap_nn.predict:run_inference:335 | Using device: cuda
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:get_keras_first_layer_channels:57 | Found first layer 'model_weights/stack0_enc0_conv0/stack0_enc0_conv0/kernel:0' with 3 input channels
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:update_backbone_in_channels:81 | Updating backbone in_channels from 1 to 3
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:map_legacy_to_pytorch_layers:272 | Successfully mapped 34/34 legacy weights to PyTorch parameters
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:365 | Successfully loaded 34/34 weights from legacy model
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:376 | Verifying weight assignments...
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:get_keras_first_layer_channels:57 | Found first layer 'model_weights/stack0_enc0_conv0/stack0_enc0_conv0/kernel:0' with 3 input channels
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:update_backbone_in_channels:81 | Updating backbone in_channels from 1 to 3
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:map_legacy_to_pytorch_layers:272 | Successfully mapped 34/34 legacy weights to PyTorch parameters
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:365 | Successfully loaded 34/34 weights from legacy model
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:376 | Verifying weight assignments...
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:get_keras_first_layer_channels:57 | Found first layer 'model_weights/stack0_enc0_conv0/stack0_enc0_conv0/kernel:0' with 3 input channels
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:update_backbone_in_channels:81 | Updating backbone in_channels from 1 to 3
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:map_legacy_to_pytorch_layers:272 | Successfully mapped 30/30 legacy weights to PyTorch parameters
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:365 | Successfully loaded 30/30 weights from legacy model
2025-10-23 11:45:22 | INFO | sleap_nn.legacy_models:load_legacy_model_weights:376 | Verifying weight assignments...
Predicting... ---------------------------------------- 100% 20/20 ETA: 0:00:00 Elapsed: 0:00:05 13.5 FPS
2025-10-23 11:45:28 | INFO | sleap_nn.predict:run_inference:453 | Finished inference at: 2025-10-23 11:45:28.862943
2025-10-23 11:45:28 | INFO | sleap_nn.predict:run_inference:454 | Total runtime: 6.7946391105651855 secs
2025-10-23 11:45:28 | INFO | sleap_nn.predict:run_inference:465 | Predictions output path: D:/SLEAP_training\predictions\20251023_SLEAPproj_redo.v001.slp.251023_114516.predictions.slp
2025-10-23 11:45:28 | INFO | sleap_nn.predict:run_inference:466 | Saved file at: 2025-10-23 11:45:28.951485

Process return code: 0

Thanks in advance!!

[sophiaanner]

This is a known bug, solution posted in Issues (VideoMatcher AUTO method incorrectly matches videos with same shape during merge #255) and quick fix mentioned in a comment on that post by ElTinmar.

[gitttt-1234]

Hi @sophiaanner !

This is fixed in the latest release, you could re-install SLEAP with the below command:

uv tool uninstall sleap

uv tool install --python 3.13 "sleap[nn]" --index https://download.pytorch.org/whl/cu128 --index https://pypi.org/simple

Thanks,

Divya