Pytorch_Retinaface
Pytorch_Retinaface copied to clipboard
biubug6
Retinaface get 80.99% in widerface hard val using mobilenet0.25.
https://github.com/pcb9382/FaceAlgorithm
直接用的训练好的pth权重跑了一下widerface，loss一直都是15左右，是训练步骤出什么问题了吗，跑了200轮关键点loss也是十几。
whats the retinaface of resnet50 input shape?
因为要放在嵌入式设备上跑，有速度要求，所以我训练了一个输入尺寸为96*96的模型，FPN只取第一层output1，anchor和step也只取第0项，数据预处理去掉了crop，别的没变。用摄像头测试时，中等脸的置信度很高，框的也准，往后退人脸变小时，框不仅不跟着变小，反而变大，人脸在里面。这是哪里的问题呢？请教各位大神，提前感谢！
I want to transfer learning using pth file (not tar file). So, in retinaface.py 57 line if cfg['name'] == 'mobilenet0.25': backbone = MobileNetV1() if cfg['pretrain']: # checkpoint = torch.load("./weights/mobilenetV1X0.25_pretrain.tar", map_location=torch.device('cpu'))...
I used `test_widerface.py` and `test_fddb.py` to test RetinaFace_ResNet50, and I found that the forward pass time became shorter and shorter with iterations. Does anyone know the reason?
Hello, I would like to ask a few questions about this github repo. 1. What do I need to do to train using the pretrained model? 2. How can I...
