orhir
about TBDC and RBDC
Hello, I’m very curious about how the calculations of TBDC and RBDC are derived. I noticed in the issues that you mentioned referring to the approach in this link , but what still confuses me is that the AED method requires tracks and anomaly regions. However, from what I see in your model, the output is only the NLL, and there doesn’t seem to be any data related to tracks or regions. Could you explain how TBDC and RBDC are computed in this case? Or perhaps I have misunderstood some aspects of your work, and you could help correct me.
Could you provide some details on how you generate the .txt file in the format of frame_id, x_min, y_min, x_max, y_max, anomaly_score to represent the detected anomaly regions from the anomaly scores and skeletal keypoints?
This data is not provided by our uploaded code by default, but it can be easily extended to include it. The additional required outputs are loaded into the dataloader; you should then post-process them and align the NLL outputs with the required data.
Hi, Thank you for your great work. I have reviewed your previous reply (No. 19) and used the attachments you provided therein, specifically the anomaly and gt files for the ShanghaiTech dataset.I performed the RBDC and TBDC calculations in the "https://github.com/lilygeorgescu/AED/tree/master/evaluation" repository, using the --num-frames parameter set to 40791, which is the total number of frames in the test set. However, the results are not ideal, with very low values of 0.009 and 0.003 respectively. I suspect there might be an error in my operation or a misunderstanding on my part. I would be extremely grateful if you could help point out where I might have gone wrong.
Thank you very much for your time and assistance.
