evaluation_README.md

Evaluation

Evaluation: Linear and FCN

For linear and FCN fine-tuning we provide a script as well. All configs can be found in experiments/linear_segmentation/configs/. An exemplary call to evaluate NeCo could look like this:

python experiments/linear_segmentation/eval_linear.py --config_path experiments/linear_segmentation/configs/coco_stuff/neco.yml

Evaluation: Overclustering

For overclustering evaluation we provide a script eval_overcluster.py under experiments/overcluster. An exemplary call to evaluate NeCo could look like this:

python experiments/overcluster/eval_overcluster.py --config_path experiments/overcluster/configs/pascal/neco.yml

Evaluation: Fully Unsupervised Semantic Segmentation

For fully unsupervised semantic segmentation, use the command below and set the arguments accordingly. For instance:

python experiments/overcluster/fully_unsup_seg.py --data_dir "" --ckpt_path ""

Evaluation: Visual In-Context Learning

Please refer to our repository where we have open-sourced the implementation of the original paper:
"Open Hummingbird Eval"
GitHub: vpariza/open-hummingbird-eval