Hand-Keypoint-Estimation

Introduction

手部21点关键点识别

TODO

• ResNet34+Finetune
• SqueezeNet+Finetune
• Hourglass
• Openpose+Design Loss
• 更好的效果展示
• 抗遮挡

Dataset

CMU手部数据集（遮挡比较变态）

Hands from Synthetic Data (6546 + 3243 + 2348 + 2124 = 14261 annotations)
└─hand_labels_synth
    ├─output_viz_synth
    ├─synth1（json文件数据缺失指尖5个关键点）
    ├─synth2
    ├─synth3
    └─synth4

Benchmarking

SqueezeNet+Finetune

Finetune = nn.Sequential(
    Flatten(), 
    nn.ReLU(),
    nn.Dropout(0.5),
    nn.Linear(247808, 256),
    #ReLU不能放BN前会导致BN方差计算错误
    nn.BatchNorm1d(256),
    nn.ReLU(),
    nn.Dropout(0.5),
    nn.Linear(256, 42),
    Reshape(-1,21,2),
    nn.Tanh()
    ) 
Total params: 64,172,906
Total trainable params: 64,172,906
Total non-trainable params: 0
Loss function : MSELoss
Epoch : 200
LR : 0.01->0.0001
Train Loss end : 0.010500	
Valid Loss end : 0.012454

CPU上0.0234s一张图片

GPU-2070Ti上0.00727s一张图片