Leonardo Araujo dos Santos

Keep going on python3 :)

What I'm trying to do.... 1. Download VOC 2007/2012, and extract at: Documents/Database/VOCdevki 2. python3 cache.py -c config.ini config/yolo/tiny-20.ini -v 3. python train.py -c config.ini config/yolo/tiny-20.ini

Hi @ruiminshen I think is working  Btw I'm documenting my findings about Yolo here: https://leonardoaraujosantos.gitbooks.io/artificial-inteligence/content/single-shot-detectors/yolo.html Do you accept a pull request for updating the Readme file?

Is because in theory you should use your own data, unfortunately I can't share the dataset.

No but I suspect it could be an issue with the dataLoader, what's your batch_size and your dataset size? I think you could try the option drop_last on the data...

Oh guys sorry for the late response, I work on this project a long time ago, and the actual data os licensed, so I can't share.

Hi @xy1999729 could you post a image or something for me to check your results? Also later on we found better ways to merge the 2 networks, I can ask...

I've added back some augmentations, Including the swap (reference, test) which is confirmed that does NOT help on the current architecture, also unfreezing the Resnet branches gave 12%+ accuracy boost...

Thanks, I will check that now

Hi @longcw now it works, here are the changes ```bash leoara01@linuxdev:~/work/yolo2-pytorch$ git diff train.py diff --git a/train.py b/train.py index 8c33cb4..51ba809 100644 --- a/train.py +++ b/train.py @@ -75,10 +75,10 @@ for...