Project of the Visual Recognition course of the Computer Vision Master
Professors: Carles Ventura and Jose Luis Gomez
Team: Group 2
- Overleaf document for the report
- Overleaf document for the FINAL REPORT
- Slides summarizing the results:
The mini-project consists on implementing in Pytorch the final classification network from M3 in order to get used to Pytorch.
- Get used to Pytorch
- Implement image classification network from M3 in Pytorch
- Use object detection models in inference
- Train Faster R-CNN on KITTI dataset
- Get familiar with KITTI-MOTS and MOTSChallenge datasets
- Use pre-trained models to evaluate the datasets
- Train Faster R-CNN and RetinaNet on the datasets
- Apply pre-trained Mask-RCNN models to KITTI-MOTS validation set
- Train Mask-RCNN model on KITTI-MOTS training set and evaluate on KITTI-MOTS validation set
- Apply pre-trained and finetuned Mask-RCNN models to MOTSChallenge training set
- Apply pre-trained and finetuned Mask-RCNN models to KITTI-MOTS validation set
- Explore and analyze the impact of different hyperparameters
- Add data augmentation techniques to Detectron2 framework
- Train your model on a synthetic dataset and finetune it on a real dataset
- Train a semantic segmentation model
- Apply tracking techniques for video object segmentation
Mini-project:
cd mini-project
python3 main.py
Week 2:
cd week2
python3 train_net.py
Week 3:
cd week3
python3 train_net.py
Week 4:
cd week4
python3 predict.py
python3 train_net.py
Week 5:
cd week4
python3 predict.py
python3 train_net.py
Week 6:
cd week5
# Data augmentation
python3 train.py
# Tracking
cd tracking
python3 test
Instructions on how to run the deeplab experiments available here.