search

vikrant7 / mobile-vod-bottleneck-lstm

Implementation of Mobile Video Object Detection with Temporally-Aware Feature Maps using PyTorch
131 stars 37 forks source link

poor MAP resualt #23

Open amiiiirrrr opened 4 years ago

amiiiirrrr commented 4 years ago

hi vikrant how did you achieved MAP=42 after two epoch? i trained mobilenet V2 + ssdlite, but almost after 20 epoch i got MAP=30. i started with lr=0.002 and schedulate it with Cosine. also, i used pre-trained mobilenet v2 classification (trained on imagenet) as pre-trained basenet and i loaded it's weights on my model.

linmoxiaomo commented 4 years ago

I trained ssd_basenet using my dataset, the loss almost unchanged. Have you met this problem?

amiiiirrrr commented 4 years ago

What was your dataset?

On Sun, 13 Sep 2020, 17:47 linmoxiaomo, notifications@github.com wrote:

I trained ssd_basenet using my dataset, the loss almost unchanged. Have you met this problem?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/vikrant7/mobile-vod-bottleneck-lstm/issues/23#issuecomment-691670372, or unsubscribe https://github.com/notifications/unsubscribe-auth/AG3PK5ZJ6FVHBBKDSQ45CUTSFTA7XANCNFSM4RJB6SDA .