Finding whether certain objects are present in images, and also where they are in the image.

CS231n Lecture 8 - Localization and Detection

http://zoey4ai.com/2018/05/12/deep-learning-object-detection/

Applications

• Face detection

Methods

Localization

–>Localization as a regression problem

• Class agnostic or class specific regressor

Localizing a fixed number of objects, has been used for human pose estimation

Sliding window detector.

video – efficient implementation

Region proposal networks from Microsoft's ResNet

Detection

Problem is that we have variable-sized outputs

Method: sliding window: Use classification on many input regions – use really fast classifier – what about for CNNs (need region proposals so that we don't have to try everywhere!)

Region proposal:

• Selective search
• many others
• EdgeBoxes

Region proposals + CNN classifiers: R-CNN, Fast R-CNNs, and Faster R-CNNs

YOLO: you only look once detection as regression. YOLO is fast!

A MultiPath Network for Object Detection (using Skip-connections)

Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks . Contextual information outside the region of interest is integrated using spatial recurrent neural networks. Inside, we use skip pooling to extract information at multiple scales and levels of abstraction – video

YOLOv3

(see here)

Image segmentation

Code

See DenseCap at Image captioning, and its references for more.

Deformable part models are CNNs?