bailvwangzi

@birdwcp I just change mining_type to NONE. So MinehardExamples() will not do hard example mining, just compute num_matches.

@zimenglan-sysu-512 i have some questions, 1.how many iterations have you trained? In the first few iterations the loss can be very large or NAN? 2.how to initialize the network? just...

@zimenglan-sysu-512 you use the two-stage method pvanet? not ssd?

@zimenglan-sysu-512 since pvanet not use OHEM, i think the 1.4% mAP boost is unfair to say focal loss is better than OHEM.

I think focal loss can be better than _softmaxWithLoss_ or _sigmoidCrossEntropyLoss_ , but it's more meaningful to prove focal loss is better than ohem, since the paper get significant improvements....

same problem. @dongxijia have you solved?

@bowenc0221 for param_group in optimizer.param_groups: new_lr = args.lr - (float(minibatch_index)*args.lr)/3e5 param_group['lr'] = new_lr

@vadim-v-lebedev 1. batchsize 1024. 2. How many epoch did you train？if you use linear decay learning rate，at the last 3e5 iterations can get better result. 3. 'use slightly less aggressive...

@chuanqi305 I test your focal loss with SSD, not mobilenet-ssd. I merge your code and change mining_type to NONE. The final loss can decrease to 0.3 , but the detection_eval...

@zhanglonghao1992 the same with you. I get up to 74 mAP after 18w iteration. To avoid the effect of initialization, I use normal SSD model(e.x. you can use normal SSD...