Skip to content

Files

Latest commit

author
Hansel
update reproducability.md
Jan 25, 2024
86391ed · Jan 25, 2024

History

History
106 lines (81 loc) · 7.16 KB

Reproducability.md

File metadata and controls

106 lines (81 loc) · 7.16 KB

Reproducability and Model weights

We provide the commands for rerunning the main experiments as well as model weights of the best models of our main experiments.

Table 1: Comparison to State-Of-The-Art

Supervised, Minimal Labels and Ours

# Ours(Opt) and Ours(OD)
python Main_Binary.py --project WSCD --seeds 42 123 7353 --annotation_time -1 --data_split train test --dataset herpes
python Main_Binary.py --seeds 42 123 7353 --max_iters 200 --classifier_lr 0.0005 --data_split test train --dataset adeno --lr_t=0.005 --lr_t_final=0.02 --nms_max_iou=0.3 --project WSCD --pseudolabels_use_validation=true  
python Main_Binary.py --seeds 42 123 7353 --max_iters 200 --classifier_lr=0.0005 --data_split test train --dataset noro --lr_t=0.005 --lr_t_final=0.02 --nms_max_iou=0.01 --project WSCD --pseudolabels_use_validation=true
python Main_Binary.py --seeds 42 123 7353 --max_iters 200 --classifier_lr=0.0005 --data_split test train --dataset papilloma --lr_t=0.001 --lr_t_final=0.02 --nms_max_iou=0.1 --project=WSCD  --pseudolabels_use_validation=true 
python Main_Binary.py --seeds 42 123 7353 --max_iters 200 --classifier_lr=0.0005 --data_split test train --dataset rota --lr_t=0.001 --lr_t_final=0.02 --nms_max_iou=0.1 --project=WSCD --pseudolabels_use_validation=true

# BB
python Main_BoundingBox.py --project WSCD --seeds 42 123 7353 --annotation_time 38027 --dataset herpes

# Loc 
python Main_Location.py --project WSCD --seeds 42 123 7353 --annotation_time 38027 --dataset herpes
python Main_Location.py --project WSCD --seeds 42 123 7353 --annotation_time 1240 --dataset adeno 
python Main_Location.py --project WSCD --seeds 42 123 7353 --annotation_time 609 --dataset noro
python Main_Location.py --project WSCD --seeds 42 123 7353 --annotation_time 587 --dataset papilloma 
python Main_Location.py --project WSCD --seeds 42 123 7353 --annotation_time 769 --dataset rota

Saliency Based Weakly Supervised Methods

  • For computing the saliency maps for GradCAM, and ViT we use pretrained classifiers and then run the ./Comparisons/ComputeCAMs.py script.

  • For TS-CAM we use the official repo, add our own datasets and save the resulting saliency maps.

  • For Reattention we use the official repo, add our own datasets and save the resulting saliency maps.

As these SOTA methods for comparison are based on saliency maps, we use ./Comparisons/Evaluate.py to compute bounding boxes from the saliency map, include the virus size and evaluate against GT labels.

Zero Shot Methods

  • For CutLer, we use the official repo to generate pseudolabels, which we then use to train a Faster RCNN.

  • For SAM, we use the huggingface implementation to generate pseudolabels, which we then use to train a Faster RCNN.

Example usage:

python Main_BoundingBox.py --project ICLR_SAM_Adeno --dataset adeno --seeds 42 123 7353 --pseudolabels sam --filter_pseudolabels true --size_range 0.6

For all compared methods we also include the known virus size. If you are interested in the implementations, please feel free to contact us.

Figure 5: Reduced Annotation Time

# Ours(Opt) and Ours(OD)
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time -1 --data_split train test 
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 28520 --data_split train test 
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 19014 --data_split train test 
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 9507 --data_split train test 
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 3803 --data_split train test 
python Main_Binary.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 1901 --data_split train test 

# Loc
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 38027
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 28520
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 19014 
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 9507
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 3803
python Main_Location.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 1901

# BB
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 38027
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 28520
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 19014 
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 9507
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 3803
python Main_BoundingBox.py --project Reduced_Annotation_Time --seeds 42 123 7353 --annotation_time 1901

Figure 6: Infinite Annotation Time

# Ours(Opt) and Ours(OD)
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage -1 --data_split train test 
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.75 --data_split train test 
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.5 --data_split train test 
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.25 --data_split train test 
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.1 --data_split train test 
python Main_Binary.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.05 --data_split train test 

# Loc
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 1
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.75
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.5 
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.25
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.1
python Main_Location.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.05

# BB
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 1
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.75
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.5 
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.25
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.1
python Main_BoundingBox.py --project Infinite_Annotation_Time --seeds 42 123 7353 --percentage 0.05