sleap-export: AttributeError: 'CentroidCropGroundTruth' object has no attribute 'keras_model'

[AgolovaUCL]

I'm trying to export a centered instance model, but I get this error, but I get the following error:

2025-01-29 18:23:12.603623: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX AVX2
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2025-01-29 18:23:12.926769: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 9403 MB memory: -> device: 0, name: NVIDIA GeForce RTX 4070 SUPER, pci bus id: 0000:01:00.0, compute capability: 8.9
Traceback (most recent call last):
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\runpy.py", line 193, in _run_module_as_main
"main", mod_spec)
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\runpy.py", line 85, in run_code
exec(code, run_globals)
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\Scripts\sleap-export.exe_main.py", line 7, in
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\site-packages\sleap\nn\inference.py", line 5012, in export_cli
max_instances=args.max_instances,
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\site-packages\sleap\nn\inference.py", line 4994, in export_model
max_instances,
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\site-packages\sleap\nn\inference.py", line 2682, in export_model
max_instances,
File "C:\Users\Sophia\AppData\Local\anaconda3\envs\sleap\lib\site-packages\sleap\nn\inference.py", line 576, in export_model
keras_model_shape = first_inference_layer.keras_model.input.shape
AttributeError: 'CentroidCropGroundTruth' object has no attribute 'keras_model'

I have tried from several computers. My sleap version is 1.4.1 and tensorflow version is 2.8.0 (although I also tried 2.7.0).

Thank you!

[gitttt-1234]

Hi @AgolovaUCL !

We won't be able to export a centered-instance model, because it needs the ground-truth centroids to crop the images. You could export a TopDown model with both centroid and centered-instance model, where the centroid model predicts the centroids of all instances in the image using which the images are cropped. The centered-instance model then predicts the keypoints in this cropped image.
You would be able to run inference (and get eval metrics) on just the centered-instance model, provided you have ground-truth data for centroids to crop the images. Sorry, we didn't explicitly mention in the sleap-export documentation, we just have it under the bonsai export guide.

Let us know if you have any questions!

Thanks,

Divya

[AgolovaUCL]

Thank you! This works. But I was wondering: sometimes when I train I only get a centroid or a centered model. What may be the cause of this?

[gitttt-1234]

Hi @AgolovaUCL !

Are you training with the GUI? It should train both centroid and centered-instance models if we select multi-animal top-down pipeline as shown below. Could you let me know how you're training the models?

Thanks,

Divya

[AgolovaUCL]

Thank you! I will try this.

…

________________________________ From: DivyaSesh ***@***.***> Sent: 30 January 2025 00:41 To: talmolab/sleap ***@***.***> Cc: Golova, Sophia ***@***.***>; Mention ***@***.***> Subject: Re: [talmolab/sleap] sleap-export: AttributeError: 'CentroidCropGroundTruth' object has no attribute 'keras_model' (Discussion #2102) ⚠ Caution: External sender Hi @AgolovaUCL<https://github.com/AgolovaUCL> ! We won't be able to export a centered-instance model, because it needs the ground-truth centroids to crop the images. You could export a TopDown model with both centroid and centered-instance model, where the centroid model predicts the centroids of all instances in the image using which the images are cropped. The centered-instance model then predicts the keypoints in this cropped image. You would be able to run inference (and get eval metrics) on just the centered-instance model, provided you have ground-truth data for centroids to crop the images. Sorry, we didn't explicitly mention in the sleap-export documentation, we just have it under the bonsai export guide<https://sleap.ai/guides/bonsai.html#top-down-model>. Let us know if you have any questions! Thanks, Divya — Reply to this email directly, view it on GitHub<#2102 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A5XPZKQY6PHUP6ZGFTHFFGL2NFYLJAVCNFSM6AAAAABWDQEZCOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMBQGE2TCOI>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[AgolovaUCL]

Hi, I believe we're resolved it! Thank you, Divya.

…

________________________________ From: DivyaSesh ***@***.***> Sent: 30 January 2025 17:00 To: talmolab/sleap ***@***.***> Cc: Golova, Sophia ***@***.***>; Mention ***@***.***> Subject: Re: [talmolab/sleap] sleap-export: AttributeError: 'CentroidCropGroundTruth' object has no attribute 'keras_model' (Discussion #2102) ⚠ Caution: External sender Hi @AgolovaUCL<https://github.com/AgolovaUCL> ! Are you training with the GUI? It should train both centroid and centered-instance models if we select multi-animal top-down pipeline as shown below. Could you let me know how you're training the models? image.png (view on web)<https://github.com/user-attachments/assets/0a480b1b-7c0c-412b-b2e5-ad909e0b5f88> Thanks, Divya — Reply to this email directly, view it on GitHub<#2102 (reply in thread)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A5XPZKV4RJPM5TTXHTHZGAT2NJLDHAVCNFSM6AAAAABWDQEZCOVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMBQHEYTINA>. You are receiving this because you were mentioned.Message ID: ***@***.***>