-
Hello @luckycallor and thanks for your implementation, i am trying to apply arc loss in my own dataset but loss drops from around 57 to around 20 and then it stops converging. Furthermore the training…
-
我在训练过程中,也是按照你脚本中设定的loss warm up =35000来设置,先使用softmax做warm up,然后使用arcloss,到34500个迭代后就挂掉,出现下面错误:
`Traceback (most recent call last):
File "train_script.py", line 269, in
train()
File "tra…
-
请问下主页链接的InsightFace_TF 里实现的arcloss 是否有了解过 我发现哪里的实现和这里的似乎有很多不一样??
-
我们想测试一下不同分辨率的识别精度问题,如果要不同的分辨率,是不是只用修改transform函数增加一个resize,还有backbone对应修改下采样就可以了。其它有和分辨率有关的要求吗?
-
I try to train a model with r50 on 'faces_emore' dataset and set --emb-size=1024;
firstly, I train a new model with softmax, and secondly, I finetune it with arcloss; but got a problem like:
mxnet.b…
-
insightFace训练过程是分两步,先做softmax的预训练再做arcLoss训练
-
https://github.com/tengshaofeng/ResidualAttentionNetwork-pytorch/blob/88ed90f1b59f4b20e152495d3a5b6a19a4aa4232/Residual-Attention-Network/train_mixup.py#L56
When testing, model do not need grad.
A…
-
In my experiments, i found that the effect of vote mode is much better than mediator. Are there some explanations for it or some tips for improving the effect of mediator.
-
-
elif args.loss_type == 4: # arc face
s = args.margin_s
m = args.margin_m
assert s > 0.0
assert m >= 0.0
assert m < (math.pi / 2)
# start to …