In mmtracking1.X, batch size only can be set as 1 in training a reid model

open-mmlab / mmtracking

OpenMMLab Video Perception Toolbox. It supports Video Object Detection (VID), Multiple Object Tracking (MOT), Single Object Tracking (SOT), Video Instance Segmentation (VIS) with a unified framework.

https://mmtracking.readthedocs.io/en/latest/

Apache License 2.0

3.58k stars 597 forks source link

In mmtracking1.X, batch size only can be set as 1 in training a reid model #913

Open heng94 opened 1 year ago

heng94 commented 1 year ago

Hi,

I am working on mmtracking 1.x and trying to train a Reid model. I found that the batch size only can be set as 1 in version 1.x.

In the old version (0.x.x), there are some processing, as shown below

But in version 1.x,

So, is there any way to improve? It's too slow to set batch size as 1 for training.