issue about generate behavior examples: no pattern image

[yochiawu]

Hi,

Thanks for your software!

When I generated the behavior examples, the program runs with the following information. But after it said the generation was completed, what I got is only one folder with all of the animation without pattern image. I am wondering is there anything wrong with that. Could you help me with this? Thanks a lot!

Preparation started... 2024-07-18 15:16:20.892583 The total categories of animals / objects in this Detector: ['mice'] The animals / objects of interest in this Detector: ['mice'] The inferencing framesize of this Detector: 480 Video fps: 30 The original video framesize: 720 X 1280 Preparation completed! Generating behavior examples... 2024-07-18 15:16:20.930974 Behavior example generation completed!

The setting is as following

[yujiahu415]

Hi, This is weird because there are two mice in the video so there should be at lease two folders '0' and '1' for storing the behavior examples. Your settings seemed all good. And the console didn't output any error message. Can you run this video in the 'Analyze Behaviors' functional unit? You don't need to select any Categorizer, just track the mice and it will output an annotated video. And you can take a look at the annotated video to see how the detection and tracking works. And also, can you provide me more information, like what's your operating system, what's the python3 version, and what's the LabGym version. Thanks!

[yochiawu]

Sorry, I didn't express clearly. Yes, there are two folders '0' and '1' storing two behavior examples animation. However, I didn't find any pattern image. I am running the 'Analyze Behaviors' now. I am using Windows system, python3.10.7,LabGym2.4.5,detectron2 0.6,CUDA12.3

Thanks!

[yujiahu415]

Great! Let me know how it goes. So you are using CUDA12.3? What is the PyTorch version you are using?

[yochiawu]

The output of 'Analyze Behaviors' looks great. It can recognize two different mice in the frame. Yes, my CUDA version is 12.3. But in the Pytorch website, I only find pytorch fits 12.1. The pytorch version is 2.3.1. And when I tested if the pytorch worked correctly, it said that the CUDA version is 12.1(but it can still use the GPU).

[yujiahu415]

Thank you for the info! This is very helpful! I didn't know Detectron2 can work with torch2.3.1 on CUDA 12.3. Back to your issue, I have been working on a new version including bug fix and it will be ready by this weekend. I'll let you know when it's ready and you can try to see if you still have this issue. Thanks again!

[yochiawu]

Great! Thanks for your time and work!

[yujiahu415]

Just had some thoughts on a possible cause of this issue: is there any special characters in the path to your videos (like in video name)? OpenCV can't write images to a path that contains non-English-letter strings.

[yochiawu]

Thanks a lot!! That's true! There is a special character in the video name. The pattern image appeared after I corrected it. Thanks again!