Topics in 3D Vision - field with many non-deep learning methods still alive and well
- Computing Correspondences
- Multi-View Stereo
- Structure from Motion
- Simultaneous Localization and Mapping (SLAM)
- Self-Supervised Learning
- View Synthesis
- Differentiable Graphics
- 3D Sensors
3D shape Prediction
![Screen Shot 2022-07-30 at 1.22.30 PM.png](https://s3-us-west-2.amazonaws.com/secure.notion-static.com/f76130aa-43ed-47ef-940c-062f9f65efc7/Screen_Shot_2022-07-30_at_1.22.30_PM.png)
- 3d image → 3d Shape
- 3d Shape → Label
3D Shape Representations
Depth Map
![Screen Shot 2022-07-30 at 1.23.06 PM.png](https://s3-us-west-2.amazonaws.com/secure.notion-static.com/ea0ee57f-5822-45b9-8828-4509ec0abb8a/Screen_Shot_2022-07-30_at_1.23.06_PM.png)
![Screen Shot 2022-07-30 at 1.23.15 PM.png](https://s3-us-west-2.amazonaws.com/secure.notion-static.com/b05ba7d2-c1fe-4258-90a0-fa6853406c15/Screen_Shot_2022-07-30_at_1.23.15_PM.png)
- Gives the Distance from the Camera to the object in the world at that pixel
- RBG image + Depth Image = RGB-D Image (2.5D)
- Problems: Cant Represent area behind objects
- Possible Task: Image → Depth Map using Fully Convolutional Network
- Per-Pixel Loss (L2 Distance) → Not quite good