- Image
- Image RoI
- Use common mask
- Use spatial encoder
- Upsample mask
- Train for 8 epochs, batch_size=24, LR=1E-4
- Optimize binary and unary predictions together after 4 epochs
- Unaries (Translation, Scale, Rotation)
- Binaries (Relative Translation, Relative Scale, Relative Direction)
nice -n 20 python -m relative3d.experiments.suncg.box3d --plot_scalars --save_epoch_freq=4 --batch_size=24 --name=box3d_base_spatial_mask_common_upsample --use_context --pred_voxels=False --classify_rot --shape_loss_wt=10 --n_data_workers=8 --num_epochs=8 --suncg_dir /nvme-scratch/nileshk/suncg/ --pred_relative=True --visdom --display_port=8094 --display_id=1 --rel_opt=True --display_freq=100 --display_visuals --auto_rel_opt=5 --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True
Finetune predictions on detections
python -m relative3d.experiments.suncg.dwr --plot_scalars --save_epoch_freq=1 --batch_size=24 --name=dwr_base_spatial_mask_common_upsample --use_context --pred_voxels=False --classify_rot --box3d_ft --box3d_pretrain_name=box3d_base_spatial_mask_common_upsample --shape_loss_wt=10 --n_data_workers=8 --num_epochs=1 --suncg_dir /nvme-scratch/nileshk/suncg/ --pred_relative=True --visdom --display_port=8094 --display_id=1 --rel_opt=True --display_freq=100 --display_visuals --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True
Fineture shape decoder
python -m relative3d.experiments.suncg.dwr --name=dwr_base_spatial_mask_common_upsample_ft --classify_rot --shape_dec_ft --use_context --plot_scalars --display_visuals --save_epoch_freq=1 --display_freq=1000 --display_id=202 --shape_loss_wt=2 --label_loss_wt=10 --batch_size=24 --num_epochs=1 --ft_pretrain_epoch=1 --ft_pretrain_name=dwr_base_spatial_mask_common_upsample --split_size=1.0 --display_port=8094 --suncg_dir=/nvme-scratch/nileshk/suncg/ --n_data_workers=4 --visdom=True --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True --pred_relative=True --rel_opt=True
Evaluate results on SunCG with GT bboxes
python -m relative3d.benchmark.suncg.box3d --num_train_epoch=4 --name=box3d_base_spatial_mask_common_upsample --classify_rot --pred_voxels=False --use_context --save_visuals --visuals_freq=50 --eval_set=val --pred_relative=True --suncg_dir=/nvme-scratch/nileshk/suncg/ --preload_stats=False --results_name=box3d_base_spatial_mask_common_upsample --do_updates=True --save_predictions_to_disk=True
Evaluate results on SunCG in detection setting to report mAP scores
python -m relative3d.benchmark.suncg.dwr --num_train_epoch=1 --name=dwr_base_spatial_mask_common_upsample --classify_rot --pred_voxels=True --use_context --save_visuals --visuals_freq=50 --eval_set=val --pred_relative=True --suncg_dir=/nvme-scratch/nileshk/suncg/ --preload_stats=False --results_name=dwr_base_spatial_mask_common_upsample --do_updates=True --save_predictions_to_disk=True --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True
We are going to fine tune the model trained on SunCG
python -m relative3d.experiments.nyu.box3d --plot_scalars --save_epoch_freq=4 --batch_size=24 --name=nyu_box3d_base_spatial_mask_common_upsample --use_context --pred_voxels=False --classify_rot --shape_loss_wt=10 --n_data_workers=8 --num_epochs=16 --nyu_dir /nfs.yoda/imisra/nileshk/nyud2/ --pred_relative=True --visdom --display_port=8094 --display_id=1 --rel_opt=True --display_freq=100 --display_visuals --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True --ft_pretrain_name=box3d_base_spatial_mask_common_upsample --ft_pretrain_epoch=8
We are going to fine tune the model trained on SunCG for detection. We do not fine tune the shape decoder on NYUv2 as the dataset has very few CAD models.
python -m relative3d.experiments.nyu.dwr --plot_scalars --save_epoch_freq=1 --batch_size=8 --name=nyu_dwr_base_spatial_mask_common_upsample --use_context --pred_voxels=False --classify_rot --ft_pretrain_name=dwr_base_spatial_mask_common_upsample --ft_pretrain_epoch=1 --shape_loss_wt=10 --n_data_workers=0 --num_epochs=16 --nyu_dir /nfs.yoda/imisra/nileshk/nyud2/ --pred_relative=True --visdom --display_port=8094 --display_id=1 --rel_opt=True --display_freq=100 --display_visuals --use_spatial_map=True --use_mask_in_common=True --upsample_mask=True