MistyRain

用仿射变换从原图切下来送进去的，代码可以参见[lib/dataset/JointsDataset.py](https://github.com/leoxiaobin/deep-high-resolution-net.pytorch/blob/ce8f3625502badde6afc6b99b5ce203f7c5caa3d/lib/dataset/JointsDataset.py#L145)

if use caffe, you can refer to [ssd-models](https://github.com/imistyrain/ssd-models)

you should add some layers without params, if use caffe, the prototxt would be ``` layer { name: "data" type: "Input" top: "data" input_param { shape: { dim: 1 dim:...

你在没有桌面环境的远程服务器运行了代码，把imshow改成imwrite之类的保存下来再拉回本地看吧

[mtcnn/detector.py](https://github.com/polarisZhao/PFLD-pytorch/blob/e438e2bae50cfb12e3c2ee6532e906c84f5660fc/mtcnn/detector.py#L287)中最小人脸大小设的只有20，如果输入图片比较大，可以根据需要调到100-300左右

方便以后推理时融合BN层加速

You can have a try of [EasyPR4Android](https://github.com/imistyrain/EasyPR4Android), which provide prebuild [apk](https://github.com/imistyrain/EasyPR4Android/releases)

[](https://stackoverflow.com/questions/52689100/data-binding-with-kotlin-processdatabinding-unable-to-get-public-no-arg-constru) 在app的build.gradle文件下的dependencies {} 块里添加: ``` // Add missing dependencies for JDK 9+ if (JavaVersion.current().ordinal() >= JavaVersion.VERSION_1_9.ordinal()) { // Add both if you're using both Kotlin and Java // If you're...

画图的代码: ``` import numpy as np import cv2 filename = "model.txt" model_points = [] raw_value = [] with open(filename) as file: for line in file: raw_value.append(line) model_points = np.array(raw_value, dtype=np.float32)...

[train.py](https://github.com/pprp/captcha.Pytorch/blob/04940e8fbd1943691d0efd09f506f20f5e181844/train.py#L344)里把net换成net = res18()就好了