This is a python script to test G-PCC. Now we only test KITTI sequences.

usage

chmod 777 tmc3 pc_error_d

python test_kitti.py --dataset='./testdata/' --mode='lossless' --prefix='lossless'
python test_kitti.py --dataset='./testdata/' --mode='lossy' --prefix='lossy'
python test_kitti.py --dataset='./testdata/' --mode='lossy' --test_d2 --prefix='lossy'
python test_kitti.py --dataset='./testdata/' --mode='lossy' --quant_mode='resolution' --test_d2 --prefix='lossy_q16bit'

Quantization

The raw KITTI point clouds are float32 numbers. In the testing, we quantize raw points to 1mm by multiplying by 1000, following the settings for Ford squences in G-PCC CTC. Other quantification methods lke 16 bits, 12 bits, etc, are also available.

Details can be found in quantize.py

G-PCC cfg

We set 'positionQuantizationScale' to [0.25, 0.125, 0.03125, 0.015625, 0.00390625, 0.001953125] to obtain 6 bitrate points, Other parameters are set to default values,

Details can be found in gpcc.py and encoder.cfg

pc error

We set resolution to 30000, following the settings for Ford squences in G-PCC CTC.

Details can be found in pc_error.py

Update

202205017: add d2 psnr; add config file.