Real time Implementations using Default Boxes

YOLO (CVPR 2016)

• Source
  • Official implementation
    • https://github.com/pjreddie/darknet
• Source Code Analysis
  • Setup for Original Paper: ./cfg/yolov1/yolo-small.cfg
  • loss function: src/detection_layer.c
    • forward_detection_layer(...)

'''python

train_yolo(...)
load_data_in_thread(...)
load_data_region(...)
fill_truth_region(...)

for (i = 0; i < count; ++i) {
    x =  boxes[i].x;
    y =  boxes[i].y;
    w =  boxes[i].w;
    h =  boxes[i].h;
    id = boxes[i].id;

    if (w < .005 || h < .005) continue;

    int col = (int)(x*num_boxes);
    int row = (int)(y*num_boxes);

    x = x*num_boxes - col;
    y = y*num_boxes - row;

    int index = (col+row*num_boxes)*(5+classes);
    if (truth[index]) continue;
    truth[index++] = 1;

    if (id < classes) truth[index+id] = 1;
    index += classes;

    truth[index++] = x;
    truth[index++] = y;
    truth[index++] = w;
    truth[index++] = h;

'''

SSD: Single Shot MultiBox Detector (ECCV 2016)

• Source

  • Official Implementation
    • https://github.com/weiliu89/caffe/tree/ssd
  • Other implementation
    • https://github.com/balancap/SSD-Tensorflow/

• Contribution points

  • Anchoring
  • Multi scale Detection (c.f. Faster-R-CNN
  • Use Convolution Layer instead of Fully Connnected Layer in the Last layer (c.f. YOLO)

• data

  • Training data: trainval.txt in VOC2012 + 2007
  • Test data: test.txt in VOC2007

Code Analysis

• example/ssd/ssd_pascal.py
  • multibox_loss_param
    • parameter를 셋팅하는 부분
• src/caffe/util/bbox_util.cpp
  • MineHardExample(...)
    • 제일 헷갈리는 부분
    • share_location 와 loc_classes: location 관련 prediction을 할때 하나의 label에 대해서 각각 따로 할 수 있고(지금까지 없었던 내용이다.) label에 상관없이 한번만 하는 방식이 있다.(지금까지 살펴본 방식)
    • Hard Negative mining을 위해서는 Highest Confidence Loss ( Class에 대한 Loss) 가 높을 수록(많이 틀릴수록) negative set을 결정한다.

YOLO v2

• Source
  • Official implementation
    • https://github.com/pjreddie/darknet
  • Other Implementation
    • https://github.com/thtrieu/darkflow

YOLO vs YOLO v2

• Anchoring
• Resolutions Final layer widthxheight: 7x7 -> 13x13 Input layer 428x428 -> 416x416
• Class predition and Bounding Box coordinates Regression detailed

Fast and Lightweight Object Detection Network: Detection and Recognition on Resource Constrained Devices

• Pontifícia Universidade Católica de Minas Gerais, 2018 ### Background

Main Strategy

• Suggest new type of labels. Grid based Segmented Label, Not Pixelwise, Not Bounding Box
• Just make heatmaps
• Shallow layers
• Reconstruct traditional Bounding Box from the heatmap