Skip to content
This repository has been archived by the owner on Nov 14, 2022. It is now read-only.

Latest commit

 

History

History
44 lines (36 loc) · 3.12 KB

kitti.md

File metadata and controls

44 lines (36 loc) · 3.12 KB
Raw

Kitti Data Experiments

Datasets

Download the KITTI raw dataset. We will use the sequences from city environments for training and evaluation.

Preprocessing

To run the evaluations, we need to compute 'dis-occluded' pixels to analyze view synthesis error over. We do so by using a (slightly modified) off the shelf stereo matching algorithm. The instructions for this preproessing are below. Note that this step is only required for evaluation.

# Download stereo matching toolbox
cd external;
wget http://ttic.uchicago.edu/~dmcallester/SPS/spsstereo.zip
unzip spsstereo.zip

# Use our patch to modify and compile
cd spsstereo;
git init; git add ./*; git commit -m “init”;
git apply ../stereo.patch
cmake .
make

# Run stereo matching
cd CODE_ROOT/lsi/data/kitti
python preprocess.py --kitti_data_root=/datasets/kitti --spss_exec=/code/lsi/external/spsstereo_git_patch/spsstereo

Training

We provide below sample scripts to train the 2 layer prediction model and the 1 layer baseline. You might need to change the paths to datasets and desired snapshot directory in the flags.

# 2 layer experiment
python ldi_enc_dec.py --dataset=kitti --kitti_data_root=/datasets/kitti --kitti_dataset_variant=raw_city --batch_size=4 --n_layers=2 --use_unet=true --num_iter=500000 --disp_smoothness_wt=0.1 --exp_name=kitti_rcity_ldi_nl2 --n_layerwise_steps=3 --trg_splat_downsampling=0.5 --compose_splat_wt=1.0 --indep_splat_wt=1.0 --self_cons_wt=10 --splat_bdry_ignore=0.05 --img_width=768 --zbuf_scale=50 --log_freq=500 --checkpoint_dir=/code/lsi/cachedir/snapshots/

# 1 layer experiment
python ldi_enc_dec.py --dataset=kitti --kitti_data_root=/datasets/kitti --kitti_dataset_variant=raw_city --batch_size=4 --n_layers=1 --use_unet=true --num_iter=500000 --disp_smoothness_wt=0.1 --exp_name=kitti_rcity_ldi_nl1 --n_layerwise_steps=3 --trg_splat_downsampling=0.5 --compose_splat_wt=1.0 --indep_splat_wt=1.0 --self_cons_wt=10 --splat_bdry_ignore=0.05 --img_width=768 --zbuf_scale=50 --log_freq=500 --checkpoint_dir=/code/lsi/cachedir/snapshots/

Evaluation

To evaluate the trained models, run:

# 2 layer experiment
python ldi_pred_eval.py  --exp_name=kitti_rcity_ldi_nl2 --train_iter=400000 --dataset=kitti --kitti_data_root=/datasets/kitti --kitti_dataset_variant=raw_city --batch_size=4 --n_layers=2 --img_width=768 --n_layerwise_steps=3 --use_unet --synth_ds_factor=2  --checkpoint_dir=/code/lsi/cachedir/snapshots --results_vis_dir=/code/lsi/cachedir/visualization/ --results_eval_dir=/code/lsi/cachedir/evaluation/  --trg_splat_downsampling=0.5 --data_split=val --num_eval_iter=250 --zbuf_scale=50 --splat_bdry_ignore=0.05

# 1 layer experiment
python ldi_pred_eval.py  --exp_name=kitti_rcity_ldi_nl1 --train_iter=400000 --dataset=kitti --kitti_data_root=/datasets/kitti --kitti_dataset_variant=raw_city --batch_size=4 --n_layers=1 --img_width=768 --n_layerwise_steps=3 --use_unet --synth_ds_factor=2  --checkpoint_dir=/code/lsi/cachedir/snapshots --results_vis_dir=/code/lsi/cachedir/visualization/ --results_eval_dir=/code/lsi/cachedir/evaluation/  --trg_splat_downsampling=0.5 --data_split=val --num_eval_iter=250 --zbuf_scale=50 --splat_bdry_ignore=0.05