search

qixuxiang / mmdetection_with_SENet154

see details in configs/retinanet_se154_fpn_1x.py
https://blog.csdn.net/xiangxianghehe/article/details/90739165
Apache License 2.0
38 stars 10 forks source link

Have you experiment on COCO with SENet？Please release the results first #1

Open nemonameless opened 5 years ago

qixuxiang commented 5 years ago

It is difficult to train senet-154 backbone on coco for my poor GPU resource (ಥ _ ಥ) The code is still under test. @nemonameless

qixuxiang commented 5 years ago

I tried senet-154 backbone retinanet on COCO dataset, I only trained 1 epoch , Here is epoch_1.pth results:

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.147
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.268
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.145
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.077
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.172
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.190
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.200
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.317
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.330
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.170
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.350
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.471

I am so sorry I am not able to train more time, anyone who have enough GPUs can try it.

You can change lr with formula lr = base_lr / 8 x num_gpus x img_per_gpu / 2,

base_lr is origin lr in config file, in retinanet base_lr = 0.01.