目标检测复习之Faster RCNN系列

目标检测之faster rcnn系列

• paper

• blogs1: 一文读懂Faster RCNN

• Faster RCNN理论合集

• code: mmdetection

Faster rcnn总结:

• 网络结构图1

• paper的图

Faster rcnn笔记总结:

• 笔记1:
  
  

ROI Aligin理解

import torch
from torchvision.ops import RoIAlign

output_size = (3, 3)
spatial_scale = 1 / 4
sampling_ratio = 2  

x = torch.FloatTensor([[
    [[1,  2,  3,  4,  5,  6 ],
     [7,  8,  9,  10, 11, 12],
     [13, 14, 15, 16, 17, 18],
     [19, 20, 21, 22, 23, 24],
     [25, 26, 27, 28, 29, 30],
     [31, 32, 33, 34, 35, 36]]
]])

rois = torch.tensor([
    [0, -2.0, -2.0, 22.0, 22.0],
])

a = RoIAlign(output_size, spatial_scale=spatial_scale, sampling_ratio=sampling_ratio)
ya = a(x, rois)
print(ya)

# 结果
tensor([[[[ 6.8333,  8.5000, 10.1667],
          [16.8333, 18.5000, 20.1667],
          [26.8333, 28.5000, 30.1667]]]])

• stackoverflow参考解释

• ROI Pooling和ROI Align