International Journal of Advanced Smart Convergence Vol.8 No.

1 65-74 (2019)
http://dx.doi.org/10.7236/IJASC.2019.8.1.65

IJASC 19-1-8

A Study on the Performance Comparison of 3D File Formats on the Web

Geon-hee Lee1, Pyeong-ho Choi2, Jeong-hwan Nam3, Hwa-seop Han3, Seung-hyun Lee1
Soon-chul Kwon1†

1
Graduate School of Smart Convergence, Kwangwoon University, Seoul, Korea
2
Department of Plasma Bio Display, Kwangwoon University, Seoul, Korea
3
MANO Landscape Design Group
1
[email protected], [email protected], [email protected],
3
[email protected],{1shlee,1†ksc0226}@kw.ac.kr

Abstract
3D file formats typically include OBJ (Wavefront file format), STL (STereoLithography), and FBX (Filmbox).
Each format has limitations depending on its configuration and usage, and supported formats are different
depending on the software application. glTF helps uniform integration of 3D file formats and allows for more
efficient transmission of large 3D geometry files by organizing them in a binary format. This paper presents
explanation on OBJ, FBX, and STL which are major examples of existing 3D file formats. It also explains the
concept and characteristics of glTF and compares its performance with other 3D file formats on the web. The
loading time and packets of each 3D file format are measured according to the web browser environment by
means of Google Chrome, Firefox and Microsoft Edge. Experimental results show that glTF is the most
efficient and that it exhibits the best performance. As to STL, relatively excessive traffic was observed. This
study is expected to contribute to reducing rendering time on the web as 3D file formats are used.

Keywords: WebGL, 3D Modeling, glTF, Comparison, Quantitative data, efficiency, STL

1. Introduction
Major 3D modeling file formats include OBJ, FBX, STL, glTF. Since such 3D file formats were
developed for different purposes, their sizes and attributes are different from one another [1]. Among these,
glTF features the multi-function support and light-weight in order for standardization of 3D file formats. In
addition, glTF can save large-size elements such as geometry in a binary format and refers to such binary
files through JSON, making the rendering process faster. This study presents details of glTF which is a
multi-functional, light-weight 3D file format in comparison of its attributes with those of other 3D file
formats being currently used such as OBJ, FBX, and STL. Additionally, the rending efficiency of each 3D
file format in a web application is analyzed based on the loading time.

Manuscript Received: Jan. 13, 2019 / Revised: Jan. 21, 2019 / Accepted: Jan. 26, 2019
Corresponding Author: [email protected]
Tel: +82-2-940-8637, Fax: +82-50-4174-3258
Graduate School of Smart Convergence, Kwangwoon University, Korea
66 International Journal of Advanced Smart Convergence Vol.8 No.1 65-74 (2019)

2. Background Theory
2.1 OBJ, FBX, STL Importer and Converter Structures
Among OBJ, FBX, and STL which are major 3D file formats, OBJ was developed by Wavefront
Technologies specifically for Visualizer. Data may be saved in an ASCII form (extension: obj) such as
DXF[2] and IGES[3] or in a binary format (extension: mod) [6]. Since it uses a separate material file entitled
MTL, MTL files that specify the texture map and material data need to be transferred as well in addition to
model information. FBX was developed by Kaydara, Autodesk has the exclusive ownership. It may be
expressed in a binary format disc or ASCII data. STL is one of the international standard formats that
represent 3-D data, and it is used commonly in an input file for most 3D printers. STL may be read with
binary or ASCII codes in a disc, but no information on model colors is saved. Figure 1 shows the process
steps of existing 3D file formats. It is impossible to apply files directly to Graphic APIs such as WebGL
immediately after such files are created by an authoring software tool such as Blender and Maya. 3D model
data shall be converted by means of the corresponding Importer and Converter depending on the Runtime
applications. The process of conversion by means of the Importer and Converter may cause a time cost to the
user.

Figure 1. Importer and Converter of OBJ, FBX, and STL

2.2 glTF Structure

The format called glTF was designed specifically for Khronos Group in order to secure efficiency
transmission of 3D contents between networks [4]. One basic element of glTF is JSON file which specifies
the structure or formation of a scene that includes 3D models. Figure 2 shows the concept of glTF Asset and
related top-level elements. Scene and Nodes are the basic structures of Scene, and Camera is an element that
sets the time point of a scene. Mesh indicates the geometric information of 3D objects, and Buffer[5], buffer
view, and accessors indicate information related to references and layouts. Materials define rendering
methods of an object. Texture, Images, and Samplers represent the outer appearance of objects [6]. Skin
indicates the peak skinning information, and Animation indicates information on time series changes. These
elements are arranged in an array format. The reference corresponding to an object is a number that indicates
a certain object in the array. This may be called an index number. A single binary glTF file includes a section
where the entire information called an Asset.
A Study on the Performance Comparison of 3D File Formats on the Web 67

Figure 2. Structure of glTF

2.3 glTF Importer and Converter Structure

Figure 3 shows the relation between the Importer and Converter as glTF is created and processed. 3D file
formats such as OBJ, FBX, and STL need to go through the process of 3D model data conversion in a form
specialized for applications that are compatible to one another so that these formats can be embodied as a
browser by means of Graphics APIs [8]. glTF needs no such conversion process or separate Importer. As a
3D model is defined by means of glTF, it is compatible with most applications that execute Graphics APIs.

Figure 3. importer and Converter of glTF(glb)

3. Experiment Environments
For an experiment, OBJ, FBX, STL, and glTF(glb) files were created by means of the Blender.
Quantitative data needs to be extracted based on the time of rendering and loading each 3D file format on the
web. In the case of OBJ, a file needs to be loaded onto a web browser by means of OBJ-MTL Loader.js. For
STL, STL Loader.js is used for rendering to a web browser. FBX needs FBX Loader.js. glTF( glb) needs
GLTF Loader.js for rendering on the web. In order to secure objective criteria of comparison, therefore, the
process of uploading by means of a loader was omitted, and a 3D model viewer supported by Autodesk was
used instead [9].
A development tool provided by Firefox, Chrome, and Microsoft Edge was used to measure the time of
rendering 3D model data on the web [10]. Such development tools indicate the time of rending on the web in
detail. Major panels frequently used for debugging include Elements, Console, Network, and Sources Panel.
68 International Journal of Advanced Smart Convergence Vol.8 No.1 65-74 (2019)

In the experiment stated above, Network Panel was used to measure the loading time of 3D model data.

Table 1. Experiment Environment

Classification Information
Google development tool
Measurement Tool Firebug
Microsoft development tool
Experimental site https://viewer.autodesk.com/
Used Browsers Firefox, Chrome, Microsoft Edge
Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz (with
Hardware
SSE4.2)
OS 64-bit Windows 8.1, build 9600
Application Dump cap (Wireshark) 2.6.4 (v2.6.4-0-g29d48ec8)
3D File format Verts Faces Tris
STL 542,969 1,084,030 1,084,030
OBJ 542,972 1,084,009 1,084,009
FBX 542,980 1,084,015 1,084,051
glTF(glb) 542,980 1,084,013 1,084,049

Table 1 shows elements used in the experiment including the browser development tool, experiment site,
browsers, hardware performance, OS, packet measuring program, and numbers of dots, faces, and triangles
of each 3D file format. Figure 4 shows OBJ, FBX, STL, and glTF(glb) in the Autodesk Viewer. Other file
formats except STL include color information. Hence, textures were applied to the rest except Figure 4(d).
Since STL has peak coordinate values but no color information, no texture is applied to it.

(a) (b)
A Study on the Performance Comparison of 3D File Formats on the Web 69

(c) (d)
Figure 4. 3D model example - (a) OBJ (b) FBX (c) glTF(glb) (d) STL

4. Experiment and Result

4.1 3D File Format Attributes
3D file formats may be often selected regardless of the use, but it is necessary to choose one depending
on the purpose since each 3D file format is developed for specific uses or purposes. Table 2 shows attributes
of each 3D file format. glTF supports important functions such as color, animation, CSG, detailed mesh
work, texture, camera, light, relative positioning, etc. ‘CSG’ stands for Constructive Solid Geometry. It uses
a boolean operator, through which a modeler can create a complex surface or object by combining simple
objects. OBJ, FBX, and STL reflect limited attributes. STL supports brief expression functions of geometry
while important functions such as color, animation, CSG, detailed mesh work, texture, camera, light, relative
positioning, etc. are omitted. OBJ supports certain functions such as brief expression, CSG, color, and
material. As mentioned earlier, OBJ uses a separate material file named MTL, and thus it is necessary to
transfer an MTL file that specifies data of the texture map or material in addition to the model information.
FBX does not provide a detailed mesh function.
Table 2. Attributes by 3D file type
Geometry Appearance Scene
File
Brief Detailed Relative Animation
Type CSG Color Material Texture Camera Light
Mesh Mesh Position
STL O X X X X X X X X X
OBJ O X O O O X X X X X
FBX O X O O O O O O O O
glTF
O O O O O O O O O O
(glb)

4.2 Performance Comparison on the Web

3D model data loading is executable in Chrome, Firefox, and Microsoft Edge browsers. It is also possible
to convert 3D model data into 3D file formats such as glTF, glb, OBJ, FBX, and STL. The loading time was
measured based on the 3 criteria: domcontentloaded, load, and Finished. The domcontentloaded event
indicates the time point when a DOM tree has been completed but an external resource (img etc..) has yet to
be loaded. The load event indicates when every resource (img, style, script etc..) has been loaded on the
70 International Journal of Advanced Smart Convergence Vol.8 No.1 65-74 (2019)

browser. ‘Finish’ indicates a time period from the beginning and end of 3D model data loading after every
source is downloaded. Table 3 shows average values after 20 times of 3D model data reloading by each of
Chrome, Firefox, and MS Edge browsers.

Table 3. Browser-specific loading time

Browser Domcontentloaded(s) Load(s) Finished(s)

glTF 2.304 6.712 36.222
glb 4.831 7.529 38.598
Chrome OBJ 5.212 12.966 41.721
FBX 2.242 14.543 43.898
STL 2.632 10.456 65.544
glTF 1.704 11.322 43.692
glb 1.812 23.534 46.53
FIrefox OBJ 1.964 23.296 46.7
FBX 1.764 22.408 44.588
STL 1.896 8.45 70.014
glTF 2.304 6.712 36.222
glb 4.831 7.529 38.598
MS Edge OBJ 5.212 12.966 41.721
FBX 2.242 14.543 43.898
STL 2.632 10.456 65.544

Figure 5(a) shows a graph of loading time in Chrome. The 3D model data loading time of glTF is shorter
than that of OBJ, FBX, and STL. That of glTF was similar to that of glb. The length of loading time was in
the order of OBJ, FBX, and STL.
Figure 5(b) shows a graph of loading time in Firefox. Just as in the case of Chrome, the 3D model data
loading time of glTF was shorter than that of OBJ, FBX, and STL in Firefox as well. However, the ranks of
OBJ and FBX in terms of loading time length were the opposite to those in Figure 5(a). This indicates that
the measurement could be different depending on the browser and network environments.
Figure 5(c) shows a graph of loading time in MS Edge. The results are almost the same with Figure 5(a).
In general, STL took a longer loading time than glTF, glb, OBJ, and FBX. Lack of RAM occurred
frequently.

(a) (b)
A Study on the Performance Comparison of 3D File Formats on the Web 71

(c)
Figure 5. Browser-specific loading time graph
(a) Chrome (b)Firefox (c)MS Edge

Since it turned in the experiment out that memory was a deciding factor affected 3D model data rendering
to a significant degree, the memory use during the rendering process was measured and is shown in Figure 6.
While the memory consumption of glTF(glb), OBJ, and FBX was the same, that of STL increased twice as
much as that in glTF (glb), OBJ, and FBX, which indicates that it is a cause of loading delay.
In order to examine the major cause, OBJ and STL were compared based on definite criteria of
comparison, and additionally, FBX, glTF, and glb were selected for comparison. The patterns of data
transmission on the network was examined by analyzing specific packets.

Figure 6. Browser-specific loading time

Figure 7(a) shows the packet segment length of 3D model data transmitted for 1 to 15 seconds the average
bit rate per second in the case of glTF. The packet segment length indicates the size of data that TCP can
bear except IP and TCP headers. The brown line indicates the average bit rate per second. Since the brown
inflection point was repeated 4 times for 15 seconds, there was a change over time depending on the quality
of the network. The blue points indicate the length of packet segments, which are 3D model data packets.
The blue line consisting of blue points indicates that packet transmission proceeded properly. At the point of
2.6 seconds where the blue line began, packet transmission was initiated. Figure 7(b) shows the length of 3D
model data packet segments transmitted for 1 to 15 seconds and the average bit rate per second in the case of
STL. At the point of 2.7 seconds where the blue line began, packet transmission was initiated. Figure 7(c)
shows the length of 3D model data packet segments transmitted for 1 to 15 seconds and the average bit rate
per second in the case of OBJ. At the point of 0.3 seconds where the blue line began, packet transmission
was initiated.
72 International Journal of Advanced Smart Convergence Vol.8 No.1 65-74 (2019)

Figure 7(d) shows the length of 3D model data packet segments transmitted for 1 to 15 seconds and the
average bit rate per second in the case of FBX. At the point of 2.2 seconds where the blue line began, packet
transmission was initiated.
Figure 7 indicates that there was a change over time depending on the quality of the network, and packet
transmission was initiated at the point where the blue line started. However, it is uncertain which 3D file
format proved a higher rate than the others even at the 4 inflection points and initial points of packet
acceptance. Thus, it was necessary to examine the beginning and end of the general packet transmission.

(a) (b)

(c) (d)
Figure 7. Packet Segments and Throughput (a) glTF(glb) (b) STL (c) OBJ (d) FBX

Figure 8(a) shows the throughput at the beginning and end of packet transmission. This experiment was
conducted in order to examine the general status at the beginning and end of packet transmission in the case
of glTF. The packet transmission of 3D model data was completed within about 27 seconds with no delay.
Figure 8(b) shows the throughput at the beginning and end of packet transmission. in the case of STL.
Delay occurred from the point of 32 seconds, and the packet transmission took more than 90 seconds. This
result indicates that while most of the 3D model data were transmitted, some part of the packet data failed to
be transmitted quickly, which resulted in a delay. Figure 8(c) shows the throughput at the beginning and end
of packet transmission. in the case of OBJ. There was no delay in packet transmission. until its completion in
25 seconds. Figure 8(d) shows the throughput at the beginning and end of packet transmission in the case of
FBX. There was no delay in 3D model data packet transmission. until its completion in about 36 seconds.
A Study on the Performance Comparison of 3D File Formats on the Web 73

(a) (b)

(c) (d)
Figure 8. Packet I/O (a) glTF (b) STL (c) OBJ (d) FBX

4. Conclusion
This study presents an overview, structure, and principles of glTF in comparison with various other 3D
file formats such as OBJ, FBX, and STL. glTF maximizes format efficiency by using JSON and binary files
to reduce the overhead and starting parsing earlier, saving large size elements such as geometry in a binary
file [18]. Since the release of glTF, Open GL, Facebook, Google, and Microsoft also have supported glTF
format for 3D model data, and it is expected that the base will be expanded gradually.
The future study needs to embody improvements of STL. A delay occurs from the point of 32 seconds
where 3D model data packet transmission is supposed to be completed, and the general transmission of
packets exceeds 90 seconds. This result indicates that while most of the 3D model data were transmitted,
some part of the packet data failed to be transmitted quickly, which resulted in a delay.
STL is used in limited areas compared to other 3D formats but advantageous in terms of user convenience
as it is applicable to THINKERCAD which is a 3D modeling website. Therefore, STL is worth more
research on functional enhancement. It is expected that this study is utilized as quantitative data useful for
users who suffer inconvenience due to rendering delays in 3D model data transmission as well as STL
researchers.

Acknowledgment
This work (Grants No. S2598342) was supported by Business for Cooperative R&D between Industry,
Academy, and Research Institute funded Ministry of SMEs and Startups in 2018.
74 International Journal of Advanced Smart Convergence Vol.8 No.1 65-74 (2019)

References

[1] Zhenhua Zhao, Kai He and Ruxu Du, "The simulation of scary robot based on OpenGL and STL," Second
International Conference on Mechanic Automation and Control Engineering, Hohhot, pp. 5429-5432, 2011.
DOI: https://doi.org/10.7236/JIIBC.2005.5.2.56.
[2] AutoCADDXF, https://en.wikipedia.org/wiki/AutoCAD_DXF
[3] IGES, https://en.wikipedia.org/wiki/IGES
[4] S. Kanamori, K. Fujiwara, T. Yoshinobu, B. Raytchev, T. Tamaki and K. Kaneda, "Physically Based Rendering of
Rainbows under Various Atmospheric Conditions," Pacific Conference on Computer Graphics and Applications,
Hangzhou, pp. 39-45, 2010.
[5] Cheng-Hsien Chen and Chen-Yi Lee, "Two-level hierarchical Z-Buffer for 3D graphics hardware," 2002 IEEE
International Symposium on Circuits and Systems. Proceedings, AZ, USA, pp. II-II, 2002.
DOI: 10.1109/ISCAS.2002.1010972
[6] Cheng-Hsien Chen and Chen-Yi Lee, "Two-level hierarchical Z-Buffer for 3D graphics hardware," 2002 IEEE
International Symposium on Circuits and Systems. Proceedings, AZ, USA, pp. II-II, 2002.
DOI: 10.1109/ISCAS.2002.1010972
[7] "IEE Colloquium on 'Binary Image Processing - Techniques and Applications' (Digest No.059)," IEE Colloquium
on Binary Image Processing - Techniques and Applications, London, UK, pp. 0-1, 1991.
[8] Geonhee Lee, Seunghyun Lee and Soonchul Kwon, "A Study on Loading Speed of Web Browser for 3D Object,"
Advanced Engineering and ICT-Convergence Proceedings, Seoul, pp. 55-56, 2018.
[9] Autodesk viewer, https://viewer.autodesk.com/designviews
[10] P. H. Shroff and S. R. Chaudhary, "Critical rendering path optimizations to reduce the web page loading time,"
International Conference for Convergence in Technology (I2CT), Mumbai, pp. 937-940, 2017.
DOI: 10.1109/I2CT.2017.8226266