Model Extraction Attack for Video Classification
Research Project For Inter Iit Tech Meet 10.0
- Performed Model Extraction of Video Swin Transformer and MoviNet trained on the Kinetics datasets in Black Box and Grey Box settings to obtain competitive results in the task
- Used conditional video generator and adverserial crafting along with knowledge distillation based techniques
- Won the gold medal in the High Prep event amongst 22 IITs across India
- Links : report; code
Papers We read
Contributor | Github Repository
- This is an open source repository maintained by Vision And Language Group containing summaries and analysis of recently published research papers
- Contributed to summaries from recent Deep Learning Conferences.
- Link : code
Semantic Segmentation using U-Net
Open Project | Vision and Language Group
- Implemented the U‑Net Architecture for semantic segmentation using PyTorch framework on the HELEN* Dataset for performing pixel‑level classification of human faces.
- Tried out various loss functions for model training and evaluated the model on class‑wise F1 scores
- Links : code