Skip to content

Latest commit

 

History

History
26 lines (23 loc) · 1.14 KB

evaluation.md

File metadata and controls

26 lines (23 loc) · 1.14 KB
Raw

Evaluation of SPICE

An example on STL10:

SPICE-Self:

 python tools/eval_self.py --config-file configs/stl10/eval.py --weight model_zoo/self_model_stl10.pth.tar --all 1

SPICE

python tools/eval_semi.py --load_path ./model_zoo/model_stl10.pth --net WideResNet_stl10 --widen_factor 2 --data_dir ./datasets/stl10 --dataset stl10 --all 1 --num_classes 10

SPICE-Self*:

 python tools/eval_self.py --config-file configs/stl10/eval.py --weight model_zoo/self_model_stl10_cls.pth.tar --all 0

SPICE*

python tools/eval_semi.py --load_path ./model_zoo/model_stl10.pth --net WideResNet_stl10 --widen_factor 2 --data_dir ./datasets/stl10 --dataset stl10 --all 0 --num_classes 10
  • Visualization of learned cluster semantics
python tools/eval_self.py --config-file configs/stl10/eval.py --weight model_zoo/self_model_stl10.pth.tar --all 1 --proto 1 --embedding ./results/stl10/embedding/feas_moco_512_l2.npy

Then, the visualization results will be saved in ./results/stl10/eval/proto/, including both the prototype examples and the corresponding localization results.