Object detection

Last updated Sep 5, 2022 Edit Source

See AI/Computer Vision/Semantic segmentation

• https://github.com/microsoft/computervision-recipes/tree/master/scenarios/detection
• #CODE MMdetection
  • OpenMMLab Detection Toolbox and Benchmark (pytorch)
  • https://mmdetection.readthedocs.io/
• #CODE TensorFlow object detection API
  • Repository
  • Model zoo
  • https://medium.com/swlh/train-your-custom-object-detector-with-tensorflow-object-detector-api-65d38dcdf08c
  • https://modelzoo.co/model/objectdetection
• #CODE Detectron2
  • Detectron2 is FAIR’s next-generation platform for object detection and segmentation
• #CODE https://github.com/jinwchoi/awesome-action-recognition#object-detection
• #CODE YOLO: Real-Time Object Detection

• #PAPER Multiple Object Recognition with Visual Attention (Ba 2015)
• #PAPER Is object localization for free? – Weakly Supervised Object Recognition with Convolutional Neural Networks (Oquab 2015):
  • http://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Oquab_Is_Object_Localization_2015_CVPR_paper.pdf
  • https://www.di.ens.fr/willow/research/weakcnn/
• #PAPER DeepLab - Weakly-and semi-supervised learning of a DCNN for semantic image segmentation (Papandreou 2015)
• #PAPER SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (Badrinarayanan 2016)
• #PAPER SSD. Single Shot MultiBox Detector (Liu 2016)
  • #BOOK https://d2l.ai/chapter_computer-vision/ssd.html
• #PAPER YOLO. You Only Look Once: Unified, Real-Time Object Detection (Redmon 2016)
  • YOLO or You Only Look Once is an object detection algorithm much different from the region based algorithms. In YOLO a single convolutional network predicts the bounding boxes and the class probabilities for these boxes.
  • How YOLO works is that we take an image and split it into an SxS grid, within each of the grid we take m bounding boxes. For each of the bounding box, the network outputs a class probability and offset values for the bounding box. The bounding boxes having the class probability above a threshold value is selected and used to locate the object within the image.
• #PAPER RetinaNet. Focal Loss for Dense Object Detection (Lin 2018)
  • #CODE https://github.com/fizyr/keras-retinanet
  • https://keras.io/examples/vision/retinanet/
  • https://towardsdatascience.com/object-detection-on-aerial-imagery-using-retinanet-626130ba2203
• #PAPER CornerNet: Detecting Objects as Paired Keypoints (Law 2018)
  • #CODE https://github.com/makalo/CornerNet
• #PAPER #REVIEW Deep learning for Generic Object Detection: A Survey (Liu 2019)
• #PAPER #REVIEW Recent Advances in Object Detection in the Age of Deep CNNs (Agarwal 2019)
• #PAPER ExtremeNet. Bottom-up Object Detection by Grouping Extreme and Center Points (Zhou 2019)
  • #CODE https://github.com/xingyizhou/ExtremeNet
• #PAPER EfficientDet: Scalable and Efficient Object Detection (Tan 2020)
  • #CODE https://github.com/xuannianz/EfficientDet
  • Included in the TF object detection API
• #CODE YOLOX: Exceeding YOLO Series in 2021 (Ge 2021)
  • #CODE https://paperswithcode.com/paper/yolox-exceeding-yolo-series-in-2021

Region-based CNNs (R-CNNs):

• https://www.pyimagesearch.com/2020/07/06/region-proposal-object-detection-with-opencv-keras-and-tensorflow/
• #BOOK Region-based RCNNs
• #PAPER Regional CNN (R-CNN)
  • The goal of R-CNN is to take in an image, and correctly identify where the main objects (via a bounding box) in the image.
  • R-CNN creates these bounding boxes, or region proposals, using a process called Selective Search.
  • Once the proposals are created, R-CNN warps the region to a standard square size and passes it through to a modified version of AlexNet (the winning submission to ImageNet 2012 that inspired R-CNN).
  • On the final layer of the CNN, R-CNN adds a Support Vector Machine (SVM) that simply classifies whether this is an object, and if so what object.
• #PAPER Fast R-CNN
  • RoI (Region of Interest) Pooling. At its core, RoIPool shares the forward pass of a CNN for an image across its subregions.
  • The second insight of Fast R-CNN is to jointly train the CNN, classifier, and bounding box regressor in a single model.
• #PAPER Faster R-CNN
  • The insight of Faster R-CNN was that region proposals depended on features of the image that were already calculated with the forward pass of the CNN (first step of classification).
  • So why not reuse those same CNN results for region proposals instead of running a separate selective search algorithm?
  • A single CNN is used to both carry out region proposals and classification. This way, only one CNN needs to be trained and we get region proposals almost for free. Faster R-CNN adds a Fully Convolutional Network on top of the features of the CNN creating what’s known as the Region Proposal Network.
• #PAPER Mask R-CNN (He 2018)
  • Extending Faster R-CNN for Pixel Level Segmentation
  • Mask R-CNN does this by adding a branch to Faster R-CNN that outputs a binary mask that says whether or not a given pixel is part of an object. The branch, as before, is just a Fully Convolutional Network on top of a CNN based feature map.
  • But the Mask R-CNN authors had to make one small adjustment to make this pipeline work as expected: Realigning RoIPool to be More Accurate.
  • https://engineering.matterport.com/splash-of-color-instance-segmentation-with-mask-r-cnn-and-tensorflow-7c761e238b46
  • https://modelzoo.co/model/mask-r-cnn-keras
  • https://machinelearningmastery.com/how-to-train-an-object-detection-model-with-keras/