SSH-TensorFlow

SSH-TensorFlow copied to clipboard

Hello, I found a Performance issue in in the definition of `train_model_random`, lib/model/train_val.py, [tf.assign](https://github.com/wanjinchang/SSH-TensorFlow/blob/6e19cbf8482510e065e5c875be9979d7b9f59412/lib/model/train_val.py#L353) will created repeatedly during program execution, resulting in reduced efficiency. I think it shoul be created...

DLPerf

Hello, I found that in the function `train_model_random`, lib/model/train_val.py, [tf.assign](https://github.com/wanjinchang/SSH-TensorFlow/blob/6e19cbf8482510e065e5c875be9979d7b9f59412/lib/model/train_val.py#L353) will create redundant node in tf computation graph repeatedly(as it is in a while loop). I think `tf.assign` should be...

JamesCao2048

@wanjinchang Hi, 感谢你的分享， 我看了代码，发现似乎不是用的标准 smooth-L1损失？是像下面这个？使用这样的 smooth-L1损失的优点是什么呢？以及它的出处是？ 

Wblossom

Hi, thanks for providing such a great repo! could you please share the pretrained models, specifically those with mobilenet backbones? Thanks

Fpooyan

the training input size

3

thanks for your jobs! I have a question about the input size for training, what size for training input you can get the wider face result you list in the...

zhouyufei0001