Queries on Non-gframe Multiple Image Processing in Training & Val.

[AngleBeats147753]

Maybe a problem has been discovered. During the training and validation processes when relevant operations are carried out, the images that are not multiples of gframe will be discarded. In the training stage, such processing may have little impact. However, for the validation stage, will it have an impact? How on earth are those images that are not multiples of gframe processed?

[Elstuhn]

My idea or thinking of how it works is that during validation stage the model goes through every frame in the video to find the bounding box and classification of each frame so every frame is processed and the images that are not multiples of gframe which are discarded are used as reference frames

But I'm not sure whether this is the case or not according to issue #113 and the author has yet to get back regarding that, I'm doing more studies and visualisations to find out in the meantime 😅

[AngleBeats147753]

Thank you very much