Pytorch_Retinaface
Pytorch_Retinaface copied to clipboard
biubug6
Retinaface get 80.99% in widerface hard val using mobilenet0.25.
When building 'bbox_overlaps', an error is raised, reminding me that `error: Unable to find a compatible Visual Studio installation`. But I have already installed VS2019 and it is added to...
I have tried to fine-tune the Resnet50_Final.pth on my own data with annotations, but the results look like the model was trained from the beginning without uploading pretrained weights. What...
Dear biubug6, Thank you very much for releasing a PyTorch implementation of RetinaFace as open source codes under the MIT license. I have a small request to you. At this...
When I train the network it trains for a few epochs then throws the following error: ``` height, width, _ = img.shape AttributeError: 'NoneType' object has no attribute 'shape' ```...
I use detect.py to detect my test image, but the result is bad. i use resnet50 Do I need to manually set the resolution of the network input? Test Image:...
The dataset link of Baidu cloud disk is invalid. Can you fix it? Thank you.
我用RetinaFace（包括在你的框架下构建的其他检测器）在不带landmark自己的数据集上训练的模型（训练size 512 x 512），发现512的测试size会发生严重的检测框偏移（都是向左下），原图测试无偏移且结果正常。但带landmark的WIDER FACE数据集训练的模型基本没有偏移，而且做landmark和不做landmark的相同检测器均存在偏移。 请问下我是哪个环节出错了，在读取label时，默认所有的gt框landmark值为-1，类别label为1，这样处理对吗。
