Link for the dataset: https://drive.google.com/drive/folders/1UqMT7axH7pZy8eydiOT90Vi_6DaNK4i-?usp=sharing
Link for the resized dataset: https://drive.google.com/drive/folders/1okE6U76brqBu-skV11iAzLYgNCX9IC1s?usp=sharing
Link to MFDETR model weights: https://drive.google.com/drive/folders/12zwQl0ijFbmulIa_fW-9_Xmx4BuGGoCg?usp=drive_link
Link to MaskDINO model weights: https://drive.google.com/drive/folders/1_4FLKe_NkUncGklEuIjqRbWVoDeJcWDy?usp=drive_link
In this repository we present the code for our ML4CV project work, focused on a litter detection and segmentation dataset (instance segmentation task).
We propose different architectures to solve the task, as well as we test some explainability methods to get some insights on the networks and their results.
The Notebooks containing our solutions are the one in bold in the following repository description.
Repository structure: Folders:
- AdelaiDet: contains libraries and code for SOLOv2
- data: contains the full annotation, the split and the resized ones for the TACO dataset
- detector: contains the code used by the TACO authors to benchmark a instance segmentation model (Mask R-CNN)
- HDDETR: contains libraries and code for Mask-Frozen DETR and DETR.
- MaskDINO: contains libraries and code for MaskDINO
- maskdino_config: contains MaskDINO config files
- res: contains model diagrams, MaskDINO training stats and test set predictions and ground truth visualization
- solov2_config: contains SOLOv2 config files
Files:
- Dataset splitting ad label replacing.ipynb: contains the script to split the official TACO dataset and get the TACO-10 subset of labels
- download.py: contains the script for the dataset downloading
- Detection based XAI.ipynb: contains the script for object detection based explainability methods
- MaskDINO Training Notebook.ipynb: contains the training script for the MaskDINO model training
- MFDETR Training Notebook.ipynb: contains the training script for the Mask-Frozen DETR model training
- MFDETR.py: contains the Mask-Frozen DETR model implementation and helper functions for loading
- README.md: this file
- requirements.txt: requirements file
- SOLOv2 Training Notebook.ipynb: contains the training script for the SOLOv2 model
- Segmentation based XAI.ipynb: containins the script for semantic segmentation based explainability methods
- validation resizing.ipynb: contains functions for the dataset resizing (images and targets) as well as for image rotation
- visual_utils.py: contains functions for segmentation-based XAI