We all are familiar with different use cases of Deep Learning among them Computer Vision is one big chunk. So, in this post, I will be covering the roadmap and topics in Computer Vision from beginner to advanced.
1. Image Classification
Given an image, need to classify what the image represents. Generally, Convolutional Neural Network is used.
For e.g A cat is sitting, it says its a cat.
 |
| Image Classification Credit: cs231n |
2. Object Localization
Given an image, locate the main object and identify the object. Here, one object is present each image.
For e.g A cat is sitting. It identifies the cat and draws a bounding box around it.
 |
| Object localization Credit: Google |
3. Object Detection
Given an image, it identifies and localize each and every object present in the image. Unlike, Object localization it can detect and identify multiple object at a time.
For e.g Two dogs and a cat is playing. It will identify two dogs and cats separately. And draw bounding box around them.
 |
| Object Detection Credit: Wikipedia |
4. Semantic Segmentation
Given an image, it masks each category of object with different colors at pixel-level
For e.g if a Pic, conatins 2 cats and 3 dogs, then the color of cats will be same and color of dogs will be same. It differentiates according to category(labels) not instance.
 |
| Semantic Segmentation Credit: Google |
5. Instance Segmentation
Unlike Semantic Segmentation, it color each instance of labelled object with different colors.
E.g If there are 3 camels and 2 person, then all the object will be colored differently at pixel-level.
It is called Instance Segmentation because it differentiates at Instance level.
 |
| Instance Segmentation Credit: Jeremy Jordan |
6. Panoptic Segmentation
This the latest improvement in the field of Segmentation. It combines the idea of Semantic Segmentation with Instance Segmentation. Because, it differentiates every object in Image like Semantic Segmentation, but also differentiates each Instances of Image from one another by masking with different color.
 |
| Panoptic Segmentation Credit: paper with code |
7. Object Tracking
It is mainly applicable in video, for a particular object it tracks the path the object is taking and also tries to project the future path of the object.
 |
| Object Detection and Tracking Credit: Python Awesome |
8. Pose Estimation
It tries to identify pose of person, by localizing the body joints and its movement. Joints here are identified as keypoints in technical term.
 |
| Pose Estimation Credit: researchgate.net |
9. 2D Image to 3D Modelling
Taking 2D flat image, it tries to recreate a 3D model from it.
 |
| 2D Image to 3D model Credit: Interesting Engineering |
10. Style Transfer
Given two images, one of normal image and image drawn by famous artists. So, it can combine, style of Artistic image with context of normal image
 |
| Style Transfer Credit: Slideshare |
No comments:
Post a Comment