Convolutional Neural Networks
R(2+1)D
Introduced by Tran et al. in A Closer Look at Spatiotemporal Convolutions for Action RecognitionA R(2+1)D convolutional neural network is a network for action recognition that employs R(2+1)D convolutions in a ResNet inspired architecture. The use of these convolutions over regular 3D Convolutions reduces computational complexity, prevents overfitting, and introduces more non-linearities that allow for a better functional relationship to be modeled.
Source: A Closer Look at Spatiotemporal Convolutions for Action Recognition
Papers
| Paper | Code | Results | Date | Stars |
|---|
Tasks
| Task | Papers | Share |
|---|---|---|
| Action Recognition | 8 | 21.62% |
| Retrieval | 4 | 10.81% |
| Video Retrieval | 4 | 10.81% |
| Temporal Action Localization | 3 | 8.11% |
| Optical Flow Estimation | 2 | 5.41% |
| Self-Supervised Action Recognition | 2 | 5.41% |
| Self-Supervised Learning | 2 | 5.41% |
| Video Recognition | 2 | 5.41% |
| Action Classification | 2 | 5.41% |
(2+1)D Convolution
Batch Normalization
Dense Connections
Global Average Pooling
ReLU
Residual Connection
