flankechen

Hi, I have meet the same fail. any quick fixed?

hi, I am training on some object detect dataset. and trying to find some guideline recall rate to stop tunning for the pnet. Can you guys report the recall of...

@victoriachen08 Pnet recall比较重要,建议建一个validation集,在validation集下recall到97%左右就可以了

@ilyanelken Hi, can you specify where you find the original author needs some "special" relu act function for Rnet and Onet? I check the paper and author's code?

@ilyanelken and I think that prelu act function is already as default in the arg scope.

@zjd1988 hi, are you going to share your code importing tflite model and run it with python? my porting to android fails.

@zjd1988 @dear-john 请问你们固化1*None*None*3作为Pnet的input之后，如何把这个.pb转化到tflite文件呢， to_tflite.sh里面的命令 `tflite_convert --output_file MTCNN_mobile/pnet2.tflite --graph_def_file frozen_graphs/pnet2.pb --input_arrays "input" --input_shapes "1,600,800,3" --output_arrays cls_prob,bbox_pred,landmark_pred --output_format TFLITE` 这个input_shapes怎么写呢， 如果我不写， 会得到这个错误 `ValueError: None is only supported in the 1st dimension. Tensor...

@zjd1988 那好像，金字塔就只能用缩放后的几个拼成一个这种做法实现了也。。。好像有点麻烦，唉

@CHANGErIBE 我是在用 [https://github.com/AITTSMD/MTCNN-Tensorflow](https://github.com/AITTSMD/MTCNN-Tensorflow) 这个基础上修改的, 3个网络的pb都可以生成. 没有试过把3个网络都freeze成一个,不过只要graph, input output names对的话,应该是可以的.

@vcvycy 嗯啊,看到了,是从facenet那个npy来的, 我在用自己训练的, 可能有点不同,谢谢了