• post_process
• • our contributions
  • files
  • environment
  • command
  • • 1.pytorch
    • 2.tensorflow
  • performance
  • contributors

post_process

Point cloud compression, point cloud post-processing, geometry refinement
This is a novel learning-based post-processing method to enhance the decompressed point clouds, which can deal with decompressed point clouds with huge variety of distortions using a single model. The source code uses pytorch as deeplearning framework, here we transplant to TensorFlow.

Paper Citation:
Xiaoqing Fan, Ge Li, Dingquan Li, Yurui Ren, Wei Gao, Thomas H. Li, “Deep Geometry Post-Processing for Decompressed Point Clouds,” IEEE International Conference on Multimedia and Expo (ICME), 2022.

our contributions

1.transplant from pytorch to TensorFlow.
2.benchmark tests on many PCs, including those not tested by author.
3.compare performances between pytorch and tensorflow.

files

1.Deep Geometry Post Processing for Decompressed Point Clouds.pdf: Paper
2.postprocess_pytorch: source code and models on pytorch
3.tensorflow: source code and models on TensorFlow
4.train-sets: go to dataset

environment

• python 3.6.9
• cuda 10.2
• pytorch 1.8.1 or tensorflow-gpu 2.3.1

command

1.pytorch

training:

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 --master_port 12345 train.py --config ./config/multi_scale.yaml --name baseline

test:

CUDA_VISIBLE_DEVICES=0 python -m torch.distributed.launch --nproc_per_node=1 --master_port 6789 test.py --config ./config/multi_scale_all.yaml --name baseline --calculate_psnr --test_list ./test_78/test_10bit.txt --single_gpu

2.tensorflow

training:
on single GPU

python train.py

on multi GPUs

python train_multiGPUs.py

test:

python test.py

performance

We perform tests on PC files from 3 publicly available datasets(for more details, pls go to: https://git.openi.org.cn/OpenPointCloud/PCC_benchmark_testsets).
For these PCs, we first carry out compression and decompression on GPCC octree and trisoup on different rates. And then impletement post-process to improve quality. From Table 1 below, we can see that post-process method works in variable compression rates and PC bit-widths of 9bit, 10bit, 11bit. It remarkably improves the quality of reconstructed PCs, except for those from lossless compression.
For D1 and D2, TensorFlow and Pytorch nearly have same performances. But Pytorch uses much more time than TensorFlow, 4-8 times. One reason is that the 3D convolution operation in TensorFlow is faster than Pytorch. Another reason is that the version in TensorFlow is baseline, as there is no 3D interpolation operators in it, which can save time but with little influence on performance.

Table 1.

file	compression	rate	ori d1_PSNR	ori d2_PSNR	TF_d1_psnr	TF_d2_psnr	TF_time	PT_d1_psnr	PT_d2_psnr	PT_time
basketball_player_vox11_00000200_dec.ply	octree	r01	52.88	55.089	59.971	60.605	32.68	59.891	60.257	194.92
basketball_player_vox11_00000200_dec.ply	octree	r02	58.814	61.529	69.285	70.69	31.62	69.716	70.85	190.69
basketball_player_vox11_00000200_dec.ply	octree	r03	64.459	68.141	76.147	78.685	32.59	76.394	78.916	203.65
basketball_player_vox11_00000200_dec.ply	octree	r04	69.234	74.351	80.956	84.325	39.62	81.078	84.226	240.14
basketball_player_vox11_00000200_dec.ply	octree	r05	74.002	79.191	83.586	87.142	59.93	83.699	87.078	279.02
basketball_player_vox11_00000200_dec.ply	octree	r06	inf	inf	67.19	67.756	59.93	64.766	65.205	291.68
basketball_player_vox11_00000200_dec.ply	trisoup	r01	59.441	64.181	53.847	54.156	52.06	51.581	51.876	307.68
basketball_player_vox11_00000200_dec.ply	trisoup	r02	67.874	71.649	59.878	60.233	53.77	54.929	55.227	434.38
basketball_player_vox11_00000200_dec.ply	trisoup	r03	72.442	75.537	64.846	65.274	55.24	57.912	58.203	418.1
basketball_player_vox11_00000200_dec.ply	trisoup	r04	73.673	76.653	67.163	67.703	58.2	61.555	61.887	420.93
dancer_vox11_00000001_dec.ply	octree	r01	52.897	55.263	59.396	59.749	27.68	59.343	59.695	263.17
dancer_vox11_00000001_dec.ply	octree	r02	58.808	61.653	68.742	69.85	27.11	69.066	70.003	382.99
dancer_vox11_00000001_dec.ply	octree	r03	64.462	68.282	75.691	78.059	28.78	75.998	77.977	533.88
dancer_vox11_00000001_dec.ply	octree	r04	69.232	74.556	80.703	83.834	34.46	80.809	83.823	608.91
dancer_vox11_00000001_dec.ply	octree	r05	74.004	79.204	83.336	86.521	42.2	83.581	86.717	644.06
dancer_vox11_00000001_dec.ply	octree	r06	inf	inf	66.823	67.356	52.83	63.996	64.411	639.12
dancer_vox11_00000001_dec.ply	trisoup	r01	59.285	63.771	54.13	54.432	47.17	51.68	51.965	361.1
dancer_vox11_00000001_dec.ply	trisoup	r02	67.644	71.633	59.763	60.092	49.5	55.324	55.626	363.53
dancer_vox11_00000001_dec.ply	trisoup	r03	72.648	75.733	64.592	64.991	49.5	58.048	58.342	277.59
dancer_vox11_00000001_dec.ply	trisoup	r04	73.965	76.896	68.043	68.64	54.63	62.281	62.624	227.42
longdress_vox10_1300_dec.ply	octree	r01	52.777	55.653	61.836	63.68	9.92	62.467	64.285	47.22
longdress_vox10_1300_dec.ply	octree	r02	58.442	62.248	68.714	71.516	9.54	69.253	71.209	54.43
longdress_vox10_1300_dec.ply	octree	r03	63.212	68.463	74.12	77.468	11.73	74.253	77.452	60.5
longdress_vox10_1300_dec.ply	octree	r04	66.771	71.931	71.086	73.826	13.29	71.03	73.619	63.44
longdress_vox10_1300_dec.ply	octree	r05	69.679	74.743	78.782	82.482	16.02	79.015	82.538	69.31
longdress_vox10_1300_dec.ply	octree	r06	inf	inf	63.304	63.996	18.3	60.211	60.682	72.55
longdress_vox10_1300_dec.ply	trisoup	r01	51.562	54.717	47.043	47.279	16.2	45.306	45.582	65.45
longdress_vox10_1300_dec.ply	trisoup	r02	59.819	61.9	52.198	52.518	16.76	49.984	50.294	65.68
longdress_vox10_1300_dec.ply	trisoup	r03	65.995	68.91	59.923	60.472	16.59	54.914	55.285	74.31
longdress_vox10_1300_dec.ply	trisoup	r04	67.904	71.038	62.066	62.65	17.73	57.415	57.823	65.83
loot_vox10_1200_dec.ply	octree	r01	52.804	55.653	66.807	69.002	9.15	68.021	71.052	46.47
loot_vox10_1200_dec.ply	octree	r02	58.439	62.217	70.8	74.162	9.34	71.305	74.727	48.3
loot_vox10_1200_dec.ply	octree	r03	63.21	68.45	74.483	77.777	11.37	74.636	77.915	61.63
loot_vox10_1200_dec.ply	octree	r04	66.775	71.923	71.068	73.711	12.36	70.989	73.476	63.33
loot_vox10_1200_dec.ply	octree	r05	69.706	74.79	79.035	82.636	15.42	79.233	82.647	78.14
loot_vox10_1200_dec.ply	octree	r06	inf	inf	61.508	62.092	17.23	58.346	58.722	74.35
loot_vox10_1200_dec.ply	trisoup	r01	51.727	56.117	46.348	46.583	15.04	44.696	44.922	69.08
loot_vox10_1200_dec.ply	trisoup	r02	60.182	64.017	50.428	50.73	15.58	48.577	48.867	71.72
loot_vox10_1200_dec.ply	trisoup	r03	65.828	69.304	57.214	57.54	15.88	52.618	52.888	70.69
loot_vox10_1200_dec.ply	trisoup	r04	67.901	71.016	60.13	60.575	17.22	55.063	55.35	71.63
queen_vox10_0200_dec.ply	octree	r01	52.805	56.04	61.371	62.129	9.55	61.804	62.589	44.77
queen_vox10_0200_dec.ply	octree	r02	58.432	62.662	68.981	71.077	9.87	69.193	71.222	54.11
queen_vox10_0200_dec.ply	octree	r03	63.21	69.048	74.051	77.041	11.9	73.789	76.518	62.68
queen_vox10_0200_dec.ply	octree	r04	66.777	72.308	71.935	74.54	13.13	71.219	73.421	66.28
queen_vox10_0200_dec.ply	octree	r05	69.707	75.013	78.375	81.762	18.51	78.737	82.179	70.25
queen_vox10_0200_dec.ply	octree	r06	inf	inf	64.489	65.252	19.46	62.451	63.011	73.96
queen_vox10_0200_dec.ply	trisoup	r01	51.64	56.169	44.75	45.029	15.99	42.608	42.885	62.5
queen_vox10_0200_dec.ply	trisoup	r02	60.665	63.629	50.969	51.289	17.41	45.857	46.133	70.9
queen_vox10_0200_dec.ply	trisoup	r03	65.763	69.252	56.914	57.256	17.4	48.849	49.11	68.48
queen_vox10_0200_dec.ply	trisoup	r04	67.568	70.877	62.553	63.082	18.54	53.147	53.413	70
redandblack_vox10_1550_dec.ply	octree	r01	52.798	55.674	60.964	62.389	8.58	61.384	62.338	49.8
redandblack_vox10_1550_dec.ply	octree	r02	58.435	62.278	67.978	70.834	9.29	68.138	70.922	54.28
redandblack_vox10_1550_dec.ply	octree	r03	63.212	68.536	73.573	76.901	10.6	73.61	76.802	56.06
redandblack_vox10_1550_dec.ply	octree	r04	66.779	71.972	71.003	73.855	11.48	70.839	73.44	60.77
redandblack_vox10_1550_dec.ply	octree	r05	69.7	74.816	78.099	81.897	14.41	78.427	82.112	64.99
redandblack_vox10_1550_dec.ply	octree	r06	inf	inf	66.609	68.088	16.22	63.359	64.182	68.79
redandblack_vox10_1550_dec.ply	trisoup	r01	49.762	54.11	45.41	46.264	13.65	44.206	44.838	63.68
redandblack_vox10_1550_dec.ply	trisoup	r02	59.033	62.101	52.067	53.037	14.28	49.695	50.539	67.42
redandblack_vox10_1550_dec.ply	trisoup	r03	65.42	68.83	58.077	59.096	14.95	53.857	54.368	66.74
redandblack_vox10_1550_dec.ply	trisoup	r04	67.868	71.17	65.541	66.728	15.9	59.014	59.497	69.32
soldier_vox10_0690_dec.ply	octree	r01	52.799	55.572	61.148	62.507	12.33	61.431	62.456	61.97
soldier_vox10_0690_dec.ply	octree	r02	58.431	62.177	68.508	71.043	12.06	68.773	71.185	67.57
soldier_vox10_0690_dec.ply	octree	r03	63.209	68.517	72.608	74.624	14.83	71.194	72.542	79.22
soldier_vox10_0690_dec.ply	octree	r04	66.773	71.935	71.156	73.939	19.13	71.045	73.691	85.34
soldier_vox10_0690_dec.ply	octree	r05	69.701	74.814	78.789	82.38	20.45	79.008	82.468	88.68
soldier_vox10_0690_dec.ply	octree	r06	inf	inf	64.707	65.696	23.03	62.119	62.757	93.81
soldier_vox10_0690_dec.ply	trisoup	r01	51.197	55.015	47.648	47.89	26.02	46.034	46.246	82.97
soldier_vox10_0690_dec.ply	trisoup	r02	60.46	62.577	53.498	53.76	20.72	50.59	50.834	88.82
soldier_vox10_0690_dec.ply	trisoup	r03	65.872	68.97	60.278	60.715	21.7	55.345	55.625	85.01
soldier_vox10_0690_dec.ply	trisoup	r04	67.952	71.171	65.195	66.161	22.92	61.645	62.17	93.17
phil_vox9_0139_dec.ply	octree	r01	46.761	50.065	54.211	55.838	6.98	54.496	56.218	21.22
phil_vox9_0139_dec.ply	octree	r02	52.409	56.843	61.611	63.989	8.3	61.025	63.838	25.18
phil_vox9_0139_dec.ply	octree	r03	57.177	63.362	66.145	69.64	8.07	66.084	69.532	24.62
phil_vox9_0139_dec.ply	octree	r04	60.742	66.298	65.409	68.521	9.28	65.333	68.408	24.83
phil_vox9_0139_dec.ply	octree	r05	63.663	68.879	69.836	74.166	9.68	70.714	74.583	28.82
phil_vox9_0139_dec.ply	octree	r06	inf	inf	59.837	61.212	11.49	56.559	57.392	29.97
phil_vox9_0139_dec.ply	trisoup	r01	43.213	51.119	37.567	38.621	8.91	35.491	36.487	21.73
phil_vox9_0139_dec.ply	trisoup	r02	51.415	57.562	43.627	44.408	10.81	38.678	39.589	26.51
phil_vox9_0139_dec.ply	trisoup	r03	57.03	62.529	49.409	50.061	9.83	42.217	43.082	27.51
phil_vox9_0139_dec.ply	trisoup	r04	61.265	65.098	57.594	58.571	9.5	50.109	50.965	28.43
phil_vox10_0139_dec.ply	octree	r01	52.794	56.27	60.819	62.216	27.14	61.143	62.728	91.28
phil_vox10_0139_dec.ply	octree	r02	58.43	63.037	67.642	70.007	28.3	67.451	69.727	99.6
phil_vox10_0139_dec.ply	octree	r03	63.208	69.655	71.899	75.27	31.52	71.907	75.238	104.03
phil_vox10_0139_dec.ply	octree	r04	66.778	72.513	71.389	74.481	35.02	71.198	74.117	114.1
phil_vox10_0139_dec.ply	octree	r05	69.696	75.042	75.642	79.469	40.49	76.265	79.961	122.14
phil_vox10_0139_dec.ply	octree	r06	inf	inf	66.762	68.317	42.96	65.934	67.325	133.69
phil_vox10_0139_dec.ply	trisoup	r01	50.011	58.232	44.364	45.693	34.83	41.352	42.428	106.65
phil_vox10_0139_dec.ply	trisoup	r02	56.32	64.292	47.569	49.052	35.94	43.341	44.408	110.4
phil_vox10_0139_dec.ply	trisoup	r03	63.55	68.619	51.957	53.463	41.75	46.616	47.639	116.45
phil_vox10_0139_dec.ply	trisoup	r04	67.181	71.228	62.054	63.252	43.07	55.297	56.221	119.81
sarah_vox9_0023_dec.ply	octree	r01	46.787	50.143	54.418	56.486	11.77	54.944	56.545	20.84
sarah_vox9_0023_dec.ply	octree	r02	52.414	56.826	61.785	64.15	13.06	61.714	64.12	21.42
sarah_vox9_0023_dec.ply	octree	r03	57.173	63.116	66.292	69.845	8.64	66.358	69.797	23.39
sarah_vox9_0023_dec.ply	octree	r04	60.749	66.179	65.492	68.596	12.45	65.376	68.325	24.83
sarah_vox9_0023_dec.ply	octree	r05	63.689	68.876	70.472	74.557	10.92	70.626	74.882	26.86
sarah_vox9_0023_dec.ply	octree	r06	inf	inf	61.853	63.88	8.98	57.771	59.106	27.85
sarah_vox9_0023_dec.ply	trisoup	r01	42.602	51.165	37.077	37.952	7.44	35.177	36.009	21.66
sarah_vox9_0023_dec.ply	trisoup	r02	47.139	58.006	40.092	41.136	8.38	36.925	37.831	23.02
sarah_vox9_0023_dec.ply	trisoup	r03	57.591	62.562	44.735	46.068	17.33	39.549	40.31	24.06
sarah_vox9_0023_dec.ply	trisoup	r04	61.059	64.699	56.53	57.93	8.69	45.462	46.064	26.85
sarah_vox10_0023_dec.ply	octree	r01	52.796	56.255	61.064	62.571	24.32	61.137	62.621	78.18
sarah_vox10_0023_dec.ply	octree	r02	58.435	63.031	67.888	70.273	24.13	67.857	70.109	82.52
sarah_vox10_0023_dec.ply	octree	r03	63.208	69.386	71.765	75.237	26.4	70.193	72.414	92.46
sarah_vox10_0023_dec.ply	octree	r04	66.775	72.338	71.473	74.476	28.79	71.327	74.237	98.15
sarah_vox10_0023_dec.ply	octree	r05	69.698	74.954	75.891	79.678	33.64	76.499	80.122	106.9
sarah_vox10_0023_dec.ply	octree	r06	inf	inf	64.525	66.232	37.32	61.153	62.355	117.61
sarah_vox10_0023_dec.ply	trisoup	r01	46.004	58.526	43.974	45.45	29.53	41.23	42.25	90.23
sarah_vox10_0023_dec.ply	trisoup	r02	56.572	64.701	45.962	47.72	32.95	42.533	43.635	100.29
sarah_vox10_0023_dec.ply	trisoup	r03	63.609	68.535	50.849	52.515	32.86	45.525	46.534	102.58
sarah_vox10_0023_dec.ply	trisoup	r04	67.138	70.862	59.747	61.086	36.69	52.614	53.527	108.36

contributors

name: Ye Hua
email: yeh@pcl.ac.cn