Cascade-FCOS

1. Standard training + evaluation

python main.py --config configs/fcos_baseline.yaml --stage 1 python main.py --config configs/refinement_k1.yaml --stage 2 --resume logs/checkpoints/stage1_epoch012.pth python main.py --config configs/refinement_k2.yaml --stage 3 --resume logs/checkpoints/stage2_epoch008.pth python evaluate.py --checkpoint logs/checkpoints/stage3_epoch006.pth --eval_stages 1 2 3 --run_name default

2. Ablation B (no retraining)

python evaluate.py --checkpoint logs/checkpoints/stage3_epoch006.pth
--eval_stages 1 2 3 --no_centerness_stages 2 3 --run_name abl_B_no_cent

3. Ablation C (retraining stages 2+3)

python main.py --config configs/ablations/abl_C_uniform_s2.yaml --stage 2 --resume logs/checkpoints/stage1_epoch012.pth python main.py --config configs/ablations/abl_C_uniform_s3.yaml --stage 3 --resume logs/checkpoints/abl_C_uniform/stage2_epoch008.pth python evaluate.py --checkpoint logs/checkpoints/abl_C_uniform/stage3_epoch006.pth
--eval_stages 1 2 3 --run_name abl_C_uniform --skip_complexity

4. Aggregate all results

python scripts/aggregate_results.py --logs_dir logs/

5. Open notebook

jupyter notebook notebooks/results_visualization.ipynb