This repo is based on 2022 Sign Spotting Challenge at ECCV.(https://chalearnlap.cvc.uab.cat/challenge/49/description/)
Papers Cited:
- Class-Balanced Loss Based on Effective Number of Samples (https://arxiv.org/abs/1901.05555v1)
- EfficientNetV2: Smaller Models and Faster Training (https://arxiv.org/abs/2104.00298)
- Densely Connected Convolutional Networks (https://arxiv.org/abs/1608.06993)
- Learning Spatiotemporal Features with 3D Convolutional Networks (https://arxiv.org/abs/1412.0767)
- 3D Convolutional Neural Networks for Human Action Recognition (https://www.dbs.ifi.lmu.de/~yu_k/icml2010_3dcnn)
- Multimodal Gesture Recognition Using 3D Convolution and Convolutional LSTM(https://ieeexplore.ieee.org/document/7880648) and some others..
->We have used PytorchVideo for processing video datasets.
->Dataset provided by the organizers uploaded at https://www.kaggle.com/datasets/gamingnation/signdataset for our use seems too much imbalanced so we prefer to use Class-Balanced Loss[1]
-Contributions:
1.Sanjay Bhandari