Presenting a Novel Pipeline for Performance Comparison of V-PCC and

G-PCC Point Cloud Compression Methods on Datasets with Varying

Properties

Albert Christensen, Daniel Lehotsk

y, Mathias Poulsen and Thomas Moeslund

Visual Analysis and Perception Lab, Aalborg University, Rendsburggade 14, DK-9000, Aalborg, Denmark

Keywords:

Point Cloud Compression, 3D Compression, V-PCC, G-PCC, Draco, Performance Comparison Pipeline.

Abstract:

The increasing availability of 3D sensors enables an ever increasing amount of applications to utilize 3D cap-

tured content in the form of point clouds. Several promising methods for compressing point clouds have been

proposed but lacks a uniﬁed method for evaluating their performance on a wide array of point cloud datasets

with different properties. We propose a pipeline for evaluating the performance of point cloud compression

methods on both static and dynamic point clouds. The proposed evaluation pipeline is used to evaluate the

performance of MPEG’s G-PCC octree RAHT and MPEG’s V-PCC compression codecs.

1 INTRODUCTION

With the increasing availability of 3D sensors such

as LiDARs, time-of-ﬂight cameras, stereo cameras

etc. more objects and scenes are captured as point

clouds. Point clouds are being used in a wide array

of applications and tasks such as autonomous driv-

ing (Li et al., 2021), virtual and augmented reality

(Bruder et al., 2014), object scanning (Chen et al.,

2016), scene scanning (Ingale and J., 2021) and ob-

ject detection and segmentation (Bello et al., 2020).

A point cloud is a simple data structure consist-

ing of a list of points containing 3D geometric in-

formation and additional attribute information such as

colours, normals, and reﬂectance, with no correlation

between the points. Since objects and scenes cap-

tured as point clouds can contain millions of points,

the storage and bandwidth requirements are often un-

feasible. Therefore, there exists a need for effective

point cloud compression codecs.

Several methods for compressing point clouds

have been proposed. Amongst these, the most no-

table are Draco by Google

, and the G-PCC (Mam-

mou et al., 2019) and V-PCC (MPEG, 2020) proposed

by the Moving Pictures Expert Group (MPEG), with

compression standards from the Joint Photographic

Expert Group (JPEG) still underway (JPEG, 2020).

In their call for proposals (MPEG, 2017), MPEG

differentiate between 3 types of point clouds - static

objects and scenes, dynamic objects, and dynamic

point cloud acquisition. While this categorization of

https://github.com/google/draco

point clouds are relevant for the use cases of MPEG’s

compression codecs, different point cloud compres-

sion methods might beneﬁt from being evaluated on a

wider set of datasets with a different set of properties,

reﬂecting other use cases.

G-PCC was made for static objects and scenes

and dynamic point cloud acquisition, while V-PCC

was made to compress dynamic objects. G-PCC, V-

PCC (Schwarz et al., 2019)(Li et al., 2020)(Kim et al.,

2020)(Liu et al., 2020) and compression codecs made

since are often evaluated on the 8iVSLF dataset (Kri-

vokuca et al., 2018), while before the standardiza-

tion efforts of MPEG, compression algorithms would

often be evaluated on completely different datasets

(Huang et al., 2008)(Fan et al., 2013)(Mekuria et al.,

2017a). Comparison between different methods are

further complicated by the datasets used for evalua-

tion of some deep learning methods for compression,

as they are often different (Quach et al., 2019)(Que

et al., 2021).

The work of (Wu et al., 2020) aimed to solve

some of these problems, by introducing their PCC

arena framework which compared different compres-

sion methods on multiple datasets. This work is lim-

ited to static datasets only and the compression of ge-

ometry only. PCC arena is the work closest to ours.

In this paper, we propose a comprehensive

pipeline for evaluating the compression performance

of codecs in a reproducible manner that allows for

easy comparison on a diverse set of publicly avail-

able point cloud datasets. Both dynamical and static

datasets are used with and without RGB attribute in-

Christensen, A., Lehotský, D., Poulsen, M. and Moeslund, T.

Presenting a Novel Pipeline for Performance Comparison of V-PCC and G-PCC Point Cloud Compression Methods on Datasets with Varying Properties.

DOI: 10.5220/0010820200003124

In Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2022) - Volume 4: VISAPP, pages

387-393

ISBN: 978-989-758-555-5; ISSN: 2184-4321

387

formation. We demonstrate the evaluation pipeline

with the V-PCC, G-PCC (RAHT / Octree) and Draco

compression codecs. The framework is modular and

can be extended with more datasets and more com-

pression codecs. Finally, we also propose an objective

metric describing the density of a point cloud as well

as a new dataset for dense static point cloud compres-

sion named MIA-Heritage.

2 METHOD

When evaluating the performance of point cloud com-

pression codecs, a set of descriptive performance met-

rics have to be chosen and evaluated on a diverse set

of benchmark datasets. We propose an evaluation

pipeline capable of comparing the compression per-

formance of various compression codecs on both dy-

namic and static datasets, with a set of quality metrics

that allows for direct comparison.

2.1 Evaluation Metrics

Evaluation metrics are needed to describe the perfor-

mance of the different compression methods. To this

end we use objective quality metrics, which describes

the quality of the reconstructed point cloud after com-

pression and decompression.

The comparison pipeline adopts the point-to-point

quality metric (D1), proposed by (Mekuria et al.,

2017b), as it is a widely used objective quality met-

ric (Schwarz et al., 2019)(Graziosi et al., 2020)(Kim

et al., 2020). D1 is the peak signal-to-noise ratio

(PSNR) reported in dB for geometry and the 3 colours

in the YCrCb colour spectrum respectively. The error

is the Euclidean distance in the geometric dimension

and the colour dimensions for the nearest neighbours

between the original and the degenerated point cloud.

Instead of using 3 different colour values, one for each

YCrCb channel as MPEG suggest, it has been cho-

sen to use the Euclidean distance in YCrCb colour

space to calculate one merged metric for the colour

PSNR. Using a single value to represent the colour

PSNR allows for easier and more direct comparison

of the codec’s performance on compressing the colour

attributes.

The Bjøntegaard-Delta (BD) metrics

(Bjøntegaard, 2001) are commonly used in the

video codec community for comparing the perfor-

mance of two different compression codecs against

each other. It is also widely used for comparing lossy

point cloud codecs at various bit rates (Gu et al.,

2019) (Santos et al., 2021) (Wang et al., 2021) (Xiong

et al., 2021). There exist two BD-metrics:

• BD-PSNR: The average PSNR difference in dB

for the same bit rate.

• BD-Rate: The average bit rate difference in per-

cent to produce the same PSNR.

The BD-metrics are found by ﬁtting the computed

D1 PSNR values for their corresponding bit rate val-

ues to a third degree logarithmic which is done for

each codec. The integral difference between codecs

can then be computed, and the average difference can

be found by dividing the integral difference over a dis-

tance. We adopt the BD-PSNR metric as the primary

evaluation metric.

The bits per input point (bpp) is chosen in favor

of the conventionally used bit rate. Bpp is a normal-

ized bit rate, making it easier to compare compression

rates of point clouds of various sizes. Bpp describes

how many bits on average that is required to represent

a single point of the input point cloud. The number

of points in the input point cloud is used instead of

the number of points in the output point cloud as the

number of points might be reduced during the com-

pression process.

2.2 Point Cloud Properties

Point clouds differ in their properties with their in-

tended usage. The compression performance of a

compression method might therefore vary between

point clouds with different properties. We propose a

set of 5 binary point cloud properties for categorizing

different point clouds.

• Type (Scene / Object)

• Density (Sparse / Dense)

• Temporality (Static / Dynamic)

• View (Single-view / Multi-view)

• Information (Geometry / Geometry + Attributes)

One element from each pair can be combined with

any other element from the following pairs. By com-

bining these characteristics, it is possible to gener-

alise point cloud types. The ﬁrst property, type, de-

scribes what is captured by the point cloud, as point

clouds can be divided into scenes or single objects

(Wu et al., 2020). A scene can be considerably more

complex than a single object. The second property,

density, describes the number of points. Dense point

clouds have a lot of points per unit area, while sparse

point clouds are relative thinly distributed. The third

property, temporality, describes whether it is a sin-

gle static frame or a dynamic point cloud sequence of

frames (Cao et al., 2019). The fourth property, view,

describes the angle from which the point cloud was

captured. Multi viewpoint clouds provide up to a 360°

VISAPP 2022 - 17th International Conference on Computer Vision Theory and Applications

388

view of the captured scene or an object. Finally, the

property information relates to the information en-

coded within the data. Geometry means that the point

cloud only consists of XYZ data, while geometry +

other attributes could be a combination of XYZ and

RGB data. However, point clouds may also include

additional information such as normals, reﬂectivity,

etc.

2.3 Surface Density

To the best of the authors knowledge the surface den-

sity property is poorly deﬁned in the literature. Thus,

we propose surface density as the metric that quanti-

ﬁes the density property of a point cloud once it has

been voxelized. Surface density for a voxelized point

cloud is deﬁned as the mean of the distance in voxels

to the nearest neighbour for each voxel in the point

cloud, see Equation (1).

∑

i=1

||v

− v

|| (1)

Where P

is the surface density of point cloud P,

n is the number of voxels in P, and v is a voxel in P,

while v

is the nearest neighbour to v

measured as

euclidian distance expressed in voxels.

Surface density of a dataset is computed by taking

the mean surface density of all point cloud sequences

in a dataset, see equation (2).

∑

i=1

(2)

Where D

is the surface density of a dataset, and S

is the mean surface density of a point cloud sequence,

with k being the number of point cloud sequences in

a dataset.

The surface density of the datasets used in this pa-

per can be seen in Figure 1.

2.4 Evaluation Dataset

The compression codecs were evaluated on the

datasets listed along with their properties in Table 1.

A total of 5 different datasets were selected in order

to have a broad range of dataset properties. The MIA-

Heritage dataset has been created for the purpose of

this paper and consist of 3 high quality scans from

the Minneapolis Institute of Art

. The dataset was

made to represent dense static objects. Additionally,

a downsampled version of MIA-Heritage is included,

where the voxelized point cloud is downsampled uni-

formly by only keeping every 50’th value in the voxel

https://github.com/HuchieWuchie/mia-heritage

Figure 1: Box plots of the computed surface densities.

Notice the small surface density in the dense datasets,

i8vfbv2, mia-heritage-high and robot@home, compared to

the sparse datasets mia-heritage-low and washington.

grid. The value of 50 was chosen by iteratively down-

sampling the point clouds more to obtain a differ-

ence in the surface density as seen in Figure 1. The

density of the downsampled dataset is considered as

sparse. This was done to compare the performance of

the compression codecs on dense and sparse datasets

where the only difference is the surface density, this

dataset is referred to as MIA-Heritage-Low. Thus, the

addition of the two MIA-heritage datasets creates a

more diverse set of point cloud properties for evalua-

tion.

The compression codecs require all datasets to be

voxelized. All datasets were voxelized to a bit depth

of 10.

Frame examples of point cloud sequences from

each dataset can be seen in Table 2.

2.5 Evaluation Pipeline

The open source evaluation pipeline

, used for evalu-

ating the different codecs with various bitrate conﬁgu-

rations on the different datasets, can be seen in Figure

The adopted metrics are calculated for all the in-

dividual point cloud sequences and averaged across

each dataset and compression codec.

It has been chosen to average the evaluation met-

rics across a whole dataset since it is not necessary

to evaluate the performance on each of the individual

point cloud sequences. This decision assumes, that

each of the datasets has similar point cloud proper-

ties.

https://github.com/math5581/PCCCP

Presenting a Novel Pipeline for Performance Comparison of V-PCC and G-PCC Point Cloud Compression Methods on Datasets with

Varying Properties

389

Table 1: Overview of the test datasets along with their properties. The MIA-Heritage-Low is downscaled by a value of 50.

Dataset Type Density Temporality View Information

8iVFB v2(d’Eon et al., 2017) Object Dense Dynamic Multi-view Geometry + RGB

Robot@Home(Ruiz-Sarmiento et al., 2017) Scene Dense Static Multi-view Geometry + RGB

MIA-Heritage

Object Dense Static Multi-view Geometry + RGB

MIA-Heritage-Low

Object Sparse Static Multi-view Geometry + RGB

Washington (Lai et al., 2011) Object Sparse Dynamic Single-view Geometry + RGB

Table 2: (a)-(c) 8iVFB-v2, (d)-(f) MIA-Heritage (dense),

(g)-(i) Robot@Home, (j)-(l) Robot@Washington RGB-D,

(m)-(o) MIA-Heritage downsampled by factor 50.

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

(m) (n) (o)

3 RESULTS

In order to evaluate the performance of several codecs

against each other using the BD metrics, an anchor is

used. The anchor serves as a baseline for comparing

the various codecs. For this report, Google’s Draco

was chosen as the anchor to compare MPEG’s G-PCC

and V-PCC against each other. Each of the tested

codecs was made to compress a speciﬁc point cloud

type. G-PCC was developed for compression of static

point cloud objects and scenes. V-PCC was developed

for compression of dynamic point cloud sequences,

such as point cloud videos. Lastly, Draco was origi-

nally developed for mesh compression. Despite this,

the comparison is made across all datasets. This is

done to evaluate their performance on the properties

deﬁned in Section 2.2.

3.1 Draco Conﬁgurations

Draco

was evaluated with the conﬁguration param-

eters given in Table 3.

Table 3: The Draco parameters that varied between the var-

ious bit rates. RX: various bitrate conﬁgurations. CL: com-

pression level, QL: quantization level.

R1 R2 R3 R4 R5 R6 R7 R8

CL 10 10 10 10 10 10 10 10

QL 6 7 8 9 10 11 12 13

3.2 G-PCC Conﬁgurations

G-PCC was evaluated with the octree transform for

geometry compression and the region adaptive Haahr

transform (RAHT) for attribute compression. Dy-

namic point clouds are compressed by compressing

the frames individually. G-PCC was evaluated with

the conﬁguration parameters given in Table 4. The

rest of the parameters were left at default. The latest

reference implementation version 14.0 was used

Table 4: The G-PCC parameters that varied between the

various bit rates. RX: various bitrate conﬁgurations. PQ:

Position quantization, LQ: Luma Quantization.

R1 R2 R3 R4 R5 R6

PQ 0.125 0.25 0.5 0.75 0.875 0.9375

LQ 51 46 40 34 28 22

3.3 V-PCC Conﬁgurations

V-PCC was evaluated with the conﬁguration param-

eters given in Table 5. Furthermore, the following

conﬁguration ﬁles where used in the written order

common/ctc-common.cfg and condition/ctc-all-

intra.cfg which can be found at the V-PCC reference

https://github.com/google/draco

https://github.com/MPEGGroup/mpeg-pcc-tmc13

VISAPP 2022 - 17th International Conference on Computer Vision Theory and Applications

390

Select Codec Select Dataset

Select Point

Cloud

Sequence

Select Point

Bitrate

Configuration

Calculate

Metrics

Draco

GPCC

VPCC

MIA-Heritage

8iVFB-v2

...

Sequence 1

Sequence 2

...

Rate 1

Rate 2

...

PSNR Geometry

PSNR Attributes

BPP

Figure 2: Illustration of the evaluation pipeline. The metrics are calculated for each combination of {Codec, Dataset/Point

Cloud Sequence and Bitrate}.

7.53

13.27

7.30

13.12

2.25

7.38

5.71

11.53

1.89

1.26

19.30

25.18

16.33

21.30

-8.95

-8.12

7.18

12.13

-7.32

2.82

-10.00

-5.00

0.00

5.00

10.00

15.00

20.00

25.00

30.00

Geometry Colour Geometry Colour Geometry Colour Geometry Colour Geometry Colour

8iVFB v2 MIA-Heritage MIA-Heritage-Low Robot@Home Washington

Gain [dB]

GPCC BD PSNR [dB] VPCC BD PSNR [dB]

Figure 3: The collected results across the different datasets, with performance difference expressed via the BD PSNR values.

The Draco codec was used as the anchor.

implementation github

. The latest reference imple-

mentation version 14.0 was used.

Table 5: The V-PCC parameters that varied between the

various bit rates. RX: various bitrate conﬁgurations. GQ:

Geometry Quantization, AQ: Attribute Quantization.

R1 R2 R3 R4

GQ 32 24 20 0

AQ 42 32 27 -12

4 DISCUSSION

It can be seen in Figure 3 that MPEG’s V-PCC out-

performs both Draco and MPEG’s G-PCC in terms

of reconstruction quality of the decompressed point

cloud for geometry and colour on the Robot@home,

8iVFB-v2 and MIA-Heritage datasets. Common to all

of these datasets is that they consist of dense point

clouds, where the distances between occupied vox-

https://github.com/MPEGGroup/mpeg-pcc-tmc2

els are low. V-PCC was designed for compressing

dense dynamic object point clouds by utilizing opti-

mized 2D video codecs to compress the colour infor-

mation, which harmonizes well with the obtained re-

sults. This also explains why the reported BD-PSNR

results for Robot@Home are lower compared to those

for 8iVFB-v2 and MIA-Heritage, as those are objects

and Robot@Home are scenes. However, V-PCC still

performs better than Draco and G-PCC in terms of

reconstruction quality on the Robot@Home dataset.

It can also be seen from Figure 3 that V-PCC com-

presses geometric information of sparse point clouds,

such as those found in the MIA-Heritage-Low and the

Washington datasets poorly, compared to G-PCC and

Draco. This is possibly because V-PCC makes use

of 2D projection methods that require a high resolu-

tion for the 2D projection planes, which is not true for

sparse point clouds.

The results on the Washington dataset suggests

that V-PCC is not only troubled by compressing

sparse point clouds but also by compressing the ge-

ometric information in single-view point clouds. This

Presenting a Novel Pipeline for Performance Comparison of V-PCC and G-PCC Point Cloud Compression Methods on Datasets with

Varying Properties

391

might be because the V-PCC codec utilizes at least 6

projection planes which are redundant for single-view

point clouds. The normal vector segmentation in the

V-PCC algorithm should have been able to take care

of this issue, however, the results suggest otherwise.

G-PCC achieves higher BD PSNR gains for geometry

which suggest that G-PCC is better suited for com-

pressing single-view point clouds.

As seen in Figure 3, MPEG’s G-PCC outperforms

both Draco and V-PCC on both MIA-Heritage-Low

and Washington in terms of both geometry and colour

reconstruction quality. For the Washington dataset

Draco performs also most identical to the G-PCC and

has a higher BD PSNR gain than V-PCC for geometry

reconstruction quality. Both the MIA-Heritage-Low

and the Washington datasets consist of sparse point

clouds where the distances between occupied voxels

are large, see Figure 1. For G-PCC the octree data

structure was chosen since it was made to compress

sparse point clouds well which seems to be conﬁrmed

by the results obtained in this paper.

It is worth noting that while G-PCC outperforms

Draco on the MIA-Heritage-Low and the Washington

datasets, it does so by a margin that is smaller than

those obtained on 8iVFB-v2 and MIA-Heritage. This

might suggest that Draco is also suitable for compres-

sion of sparse point clouds. Both Draco and G-PCC

outperform V-PCC heavily when compressing geo-

metric information in sparse point clouds. Further-

more, Figure 3 shows that G-PCC performs well on a

wide variety of datasets and does so consistently, e.g.

G-PCC performs almost similar on dense object and

dense scene datasets.

5 CONCLUSION

This paper proposes a set of binary properties to de-

scribe point clouds and argues that point cloud com-

pression methods should be evaluated on a diverse set

of datasets with different properties. To this end, an

evaluation framework with an associated open source

evaluation pipeline has been proposed with publicly

available datasets.

Furthermore we also propose the MIA-Heritage

dataset as a static dense point cloud compression

dataset benchmark, as well as a metric for surface

density to evaluate whether a point cloud is sparse or

dense.

The evaluation of 3D compression methods ﬁnds

that V-PCC provides good reconstruction quality on

dense static and dense dynamic point clouds. It per-

forms the strongest on objects but also outperforms

Draco and G-PCC on dense scenes in terms of recon-

struction quality. V-PCC is outperformed by G-PCC

and Draco on sparse datasets, with Draco and G-PCC

performing somewhat equally. Furthermore, the re-

sults suggest that V-PCC is challenged on single-view

datasets in terms of geometric reconstruction quality.

REFERENCES

Bello, S. A., Yu, S., and Wang, C. (2020). Review: deep

learning on 3d point clouds. CoRR, abs/2001.06280.

Bjøntegaard, G. (2001). Calculation of average psnr differ-

ences between rd-curves.

Bruder, G., Steinicke, F., and N

uchter, A. (2014). Poster:

Immersive point cloud virtual environments. In 2014

IEEE Symposium on 3D User Interfaces (3DUI),

pages 161–162.

Cao, C., Preda, M., and Zaharia, T. (2019). 3d point cloud

compression: A survey. In The 24th International

Conference on 3D Web Technology, pages 1–9.

Chen, L.-C., Hoang, D.-C., Lin, H.-I., and Nguyen, T.-H.

(2016). Innovative methodology for multi-view point

cloud registration in robotic 3d object scanning and

reconstruction. Applied Sciences, 6(5).

d’Eon, E., Harrison, B., Myers, T., and Chou,

P. A. (2017). 8i voxelized full bodies-a vox-

elized point cloud dataset. ISO/IEC JTC1/SC29

Joint WG11/WG1 (MPEG/JPEG) input document

WG11M40059/WG1M74006, 7:8.

Fan, Y., Huang, Y., and Peng, J. (2013). Point cloud com-

pression based on hierarchical point clustering. In

2013 Asia-Paciﬁc Signal and Information Processing

Association Annual Summit and Conference, pages 1–

Graziosi, D., Nakagami, O., Kuma, S., Zaghetto, A.,

Suzuki, T., and Tabatabai, A. (2020). An overview of

ongoing point cloud compression standardization ac-

tivities: video-based (v-pcc) and geometry-based (g-

pcc). APSIPA Transactions on Signal and Information

Processing, 9:e13.

Gu, S., Hou, J., Zeng, H., Yuan, H., and Ma, K.-K. (2019).

3d point cloud attribute compression using geometry-

guided sparse representation. IEEE Transactions on

Image Processing, 29:796–808.

Huang, Y., Peng, J., Kuo, C.-C. J., and Gopi, M. (2008).

A generic scheme for progressive point cloud cod-

ing. IEEE Transactions on Visualization and Com-

puter Graphics, 14(2):440–453.

Ingale, A. K. and J., D. U. (2021). Real-time 3d reconstruc-

tion techniques applied in dynamic scenes: A sys-

tematic literature review. Computer Science Review,

39:100338.

JPEG (2020). Final call for evidence on jpeg pleno point

cloud coding. ISO/IEC JTC 1/SC 29/WG 1 (ITU-T

SG16).

Kim, J., Im, J., Rhyu, S., and Kim, K. (2020). 3d motion

estimation and compensation method for video-based

point cloud compression. IEEE Access, 8:83538–

83547.

VISAPP 2022 - 17th International Conference on Computer Vision Theory and Applications

392

Krivokuca, M., Chou, P. A., and Savill, P. (2018). 8i vox-

elized surface light ﬁeld (8ivslf) dataset. ISO/IEC

JTC1/SC29/WG11 MPEG, input document m42914.

Lai, K., Bo, L., Ren, X., and Fox, D. (2011). A large-

scale hierarchical multi-view rgb-d object dataset. In

2011 IEEE international conference on robotics and

automation, pages 1817–1824. IEEE.

Li, L., Li, Z., Zakharchenko, V., Chen, J., and Li, H. (2020).

Advanced 3d motion prediction for video-based dy-

namic point cloud compression. IEEE Transactions

on Image Processing, 29:289–302.

Li, Y., Ma, L., Zhong, Z., Liu, F., Chapman, M. A., Cao,

D., and Li, J. (2021). Deep learning for lidar point

clouds in autonomous driving: A review. IEEE Trans-

actions on Neural Networks and Learning Systems,

32(8):3412–3432.

Liu, H., Yuan, H., Liu, Q., Hou, J., and Liu, J. (2020).

A comprehensive study and comparison of core tech-

nologies for mpeg 3-d point cloud compression. IEEE

Transactions on Broadcasting, 66(3):701–717.

Mammou, K., Chou, P. A., Flynn, D., Krivoku

ca, M., Nak-

agami, O., and Sugio, T. (2019). G-pcc codec descrip-

tion v2. ISO/IEC JTC1/SC29/WG11 N18189.

Mekuria, R., Blom, K., and Cesar, P. (2017a). Design, im-

plementation, and evaluation of a point cloud codec

for tele-immersive video. IEEE Transactions on Cir-

cuits and Systems for Video Technology, 27(4):828–

842.

Mekuria, R., Laserre, S., and Tulvan, C. (2017b). Perfor-

mance assessment of point cloud compression. In

2017 IEEE Visual Communications and Image Pro-

cessing (VCIP), pages 1–4.

MPEG (2017). Call for proposals for point cloud

compression v2. ISO/IEC JTC1/SC29/WG11

MPEG2017/N16763.

MPEG (2020). V-pcc codec description. ISO/IEC JTC 1/SC

29/WG 7.

Quach, M., Valenzise, G., and Dufaux, F. (2019). Learning

convolutional transforms for lossy point cloud geome-

try compression. 2019 IEEE International Conference

on Image Processing (ICIP).

Que, Z., Lu, G., and Xu, D. (2021). Voxelcontext-net: An

octree based framework for point cloud compression.

Ruiz-Sarmiento, J., Galindo, C., and Gonzalez-Jimenez, J.

(2017). Robot@home, a robotic dataset for seman-

tic mapping of home environments. The International

Journal of Robotics Research, 36(2):131–141.

Santos, C., Gonc¸alves, M., Corr

ea, G., and Porto, M.

(2021). Block-based inter-frame prediction for dy-

namic point cloud compression. In 2021 IEEE In-

ternational Conference on Image Processing (ICIP),

pages 3388–3392. IEEE.

Schwarz, S., Preda, M., Baroncini, V., Budagavi, M.,

Cesar, P., Chou, P. A., Cohen, R. A., Krivoku

ca,

M., Lasserre, S., Li, Z., Llach, J., Mammou, K.,

Mekuria, R., Nakagami, O., Siahaan, E., Tabatabai,

A., Tourapis, A. M., and Zakharchenko, V. (2019).

Emerging mpeg standards for point cloud compres-

sion. IEEE Journal on Emerging and Selected Topics

in Circuits and Systems, 9(1):133–148.

Wang, J., Ding, D., Li, Z., and Ma, Z. (2021). Multiscale

point cloud geometry compression. In 2021 Data

Compression Conference (DCC), pages 73–82. IEEE.

Wu, C.-H., Hsu, C.-F., Kuo, T.-C., Griwodz, C., Riegler,

M., Morin, G., and Hsu, C.-H. (2020). Pcc arena: a

benchmark platform for point cloud compression al-

gorithms. Proceedings of the 12th ACM International

Workshop on Immersive Mixed and Virtual Environ-

ment Systems.

Xiong, J., Gao, H., Wang, M., Li, H., and Lin, W. (2021).

Occupancy map guided fast video-based dynamic

point cloud coding. IEEE Transactions on Circuits

and Systems for Video Technology.

Presenting a Novel Pipeline for Performance Comparison of V-PCC and G-PCC Point Cloud Compression Methods on Datasets with

Varying Properties

393