Open nabeelkhalid92 opened 6 months ago
Hello.
Due to the fast FPS of the image sequence and the small distance of object movement between frames, some frames were skipped when creating the object detection dataset.
In addition, if some categories have a large proportion, they will be skipped more to balance the categories.
For specific code reference: https://github.com/ucsk/ppbytetrack/blob/main/tools/ctmc2voc.py#L98-L99.
Thank you very much for your kind reply!
Could you please explain what this image represents?
Additionally, I noticed that for PPByteTrack, you used the last 75% of the data for training and the first 25% for testing. For PPBotSort, you did the opposite, using the first 25% for training and the last 75% for testing. Is this correct?
The code you provided currently splits the data using the last 75% for training and the first 25% for validation, correct? If I want to perform the same data split as you did for PPBotSort (first 75% for training and last 25% for validation), what changes do I need to make to the code?
I have a question related to the data sampling. Which parameters or predefined criteria did you use to reduce the total number of images?
Thank you