TRAIN.md

Use the commands below to train PARSAC according to the experiments in our paper.

We use Weights & Biases to log the training progress. You can enable it with --wandb online to use online syncing or --wandb offline to save the logs in a folder offline. Use the options --wandb_entity, --wandb_group and --wandb_dir to set the entity, group and local directory for your logs.

Main Results

Vanishing Points

SU3

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp --instances 8 --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine 

NYU-VP

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset nyuvp --problem vp --instances 8 --hypsamples 64 --data_path datasets/nyu_vp/data --checkpoint_dir ./tmp/checkpoints --no_refine 

Fundamental Matrices

HOPE-F

python parsac.py --hypotheses 32 --batch 32 --samplecount 16 --inlier_threshold 0.004 --assignment_threshold 0.02 --dataset hope --problem fundamental --instances 4 --hypsamples 128 --epochs 3000 --lr_steps 2500 --data_path datasets/hope --checkpoint_dir ./tmp/checkpoints

Homographies

SMH

python parsac.py --hypotheses 32 --batch 4 --samplecount 8 --inlier_threshold 1e-6 --assignment_threshold 4e-6 --dataset smh --problem homography --instances 24 --hypsamples 64 --epochs 500 --lr_steps 350 --data_path datasets/smh --checkpoint_dir ./tmp/checkpoints

Self-Supervised Learning

Weighted Loss

SU3

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp --instances 8 --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine --self_supervised

NYU-VP

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset nyuvp --problem vp --instances 8 --hypsamples 64 --data_path datasets/nyu_vp/data --checkpoint_dir ./tmp/checkpoints --no_refine --self_supervised

HOPE-F

python parsac.py --hypotheses 32 --batch 32 --samplecount 16 --inlier_threshold 0.004 --assignment_threshold 0.02 --dataset hope --problem fundamental --instances 4 --hypsamples 128 --epochs 3000 --lr_steps 2500 --data_path datasets/hope --checkpoint_dir ./tmp/checkpoints --self_supervised

Unweighted Loss

SU3

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp --instances 8 --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine --self_supervised --cumulative_loss -1

NYU-VP

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset nyuvp --problem vp --instances 8 --hypsamples 64 --data_path datasets/nyu_vp/data --checkpoint_dir ./tmp/checkpoints --no_refine --self_supervised --cumulative_loss -1

HOPE-F

python parsac.py --hypotheses 32 --batch 32 --samplecount 16 --inlier_threshold 0.004 --assignment_threshold 0.02 --dataset hope --problem fundamental --instances 4 --hypsamples 128 --epochs 3000 --lr_steps 2500 --data_path datasets/hope --checkpoint_dir ./tmp/checkpoints --self_supervised --cumulative_loss -1

Ablation Study: Number of Model Instances

Replace M with the desired number of putative model instances:

python parsac.py --instances M --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp  --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine 

Ablation Study: (Un-)weighted Inlier Counting

w/o weighted inlier counting

SU3

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp --instances 8 --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine --inlier_counting unweighted

HOPE-F

python parsac.py --hypotheses 32 --batch 32 --samplecount 16 --inlier_threshold 0.004 --assignment_threshold 0.02 --dataset hope --problem fundamental --instances 4 --hypsamples 128 --epochs 3000 --lr_steps 2500 --data_path datasets/hope --checkpoint_dir ./tmp/checkpoints --inlier_counting unweighted

SMH

python parsac.py --hypotheses 32 --batch 4 --samplecount 8 --inlier_threshold 1e-6 --assignment_threshold 4e-6 --dataset smh --problem homography --instances 24 --hypsamples 64 --epochs 500 --lr_steps 350 --data_path datasets/smh --checkpoint_dir ./tmp/checkpoints --inlier_counting unweighted

Ablation Study: Feature Generalisation

Train on DeepLSD

SU3

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset su3 --problem vp --instances 8 --hypsamples 64 --data_path datasets/su3 --checkpoint_dir ./tmp/checkpoints --no_refine --ablation_deeplsd_folder deeplsd_features/su3

NYU-VP

python parsac.py --hypotheses 32 --batch 64 --samplecount 8 --inlier_threshold 0.0001 --dataset nyuvp --problem vp --instances 8 --hypsamples 64 --data_path datasets/nyu_vp/data --checkpoint_dir ./tmp/checkpoints --no_refine --ablation_deeplsd_folder deeplsd_features/nyu