This repository expects the following external assets to be placed in the corresponding folders before running the experiments.
Place the MMDetection Faster R-CNN checkpoint in:
checkpoints/
Expected checkpoint:
faster_rcnn_x101_32x4d_fpn_mstrain_3x_coco
This checkpoint corresponds to the MMDetection Faster R-CNN X101-32x4d-FPN model trained on COCO.
Place the COCO 2017 dataset in:
datasets/
Expected dataset folder:
datasets/coco-2017/
Place the detector weights in:
detectors/
Expected detector weights:
detectors/yolo26x
detectors/rtdetr-l
The runs/ folder contains the pretrained SCOPE weights used for the experiments.
These pretrained weights can be downloaded from Google Drive:
https://drive.google.com/drive/folders/1LFbt_ZrvPJgfbSr174lVPzxeQEkMO0Uz?usp=sharing
After downloading, place the pretrained run folders inside:
runs/
A minimal expected layout is:
.
├── checkpoints/
│ └── faster_rcnn_x101_32x4d_fpn_mstrain_3x_coco
├── datasets/
│ └── coco-2017/
├── detectors/
│ ├── yolo26x
│ └── rtdetr-l
├── runs/
│ └── <downloaded_pretrained_scope_weights>
└── ...
Large assets such as datasets, checkpoints, pretrained weights, generated outputs, and plots are not meant to be committed to Git.