• Grand Challenge on 360-degree Video on-demand Streaming

Organized and sponsored by

Challenge Description

In recent times, 360-degree video on-demand streaming has gathered significant interest across academia and the industry. With sales of VR headsets from prominent brands like Meta and Pico surpassing 10 million units and Apple’s recent entry into the scene with its debut spatial computing device, it is evident that the horizon for 360-degree videos and their applications is expanding rapidly. These videos, distinct from traditional 2D counterparts, come with challenges like immense data transmission volume and stringent interactive latency demands. Due to the complexity of the overall system, there is yet no universally endorsed processing solution for 360-degree video transmission.

To foster advancements in 360-degree video on-demand streaming, ByteDance presents this challenge. Participants will benefit from our open-source evaluation platform, E3POexternal link, designed entirely in Python. E3PO facilitates the simulation and assessment of a 360-degree video streaming system, empowering contestants to focus on the creation and finesse of pivotal algorithms.


Contestants shall design and implement a 360-degree video on-demand streaming solution using E3PO. The goal of the solution is to deliver the best user viewing quality using the least system resources. We use the objective video quality of user’s actual viewing area on the terminal device, measured by $MSE$ to evaluate the user viewing quality. In terms of system recources, we consider three major costs:

  • $w_1 * C_b$: the bandwidth cost of streaming all data from the server to user
  • $w_2*C_s$: the storage cost of storing video data on the server
  • $w_3*C_c$: the computation cost for some solutions that require real-time processing or transcoding

We define a performance metric score as follows. The denominator of the formula can be considered as a Lagrangian variant of the rate-distortion optimization problem, and its physical interpretation is to minimize the distortion and cost simultaneously.

$$ S = \frac{1}{ \alpha * MSE + \beta * (w_1 * C_b + w_2 * C_s + w_3 * C_c)} $$

The weight coefficients $\alpha = 0.006$ and $\beta=10$ ($\alpha$ and $\beta$ may changes for different test video sequences) are used to control the distortion and cost, respectively. Meanwhile, for the weights $w_1=0.09$, $w_2=0.000015$, and $w_3=0.000334$ in the cost, we referred to the pricing table on the AWSexternal link official website. The unit for $C_b$ and $C_s$ is GB, and $C_c$ represents the duration of the video playback in seconds.

Note that E3PO automatically measures performance metrics and calculates $S$ for each simulation. We provide 8K panoramic video sequences as well as real users’ head motion traceexternal link data for contestants’ testing and final evaluation.

Important Dates

Please refer to the important dates page.


This challenge is open to any individual, academic or commercial institution (except ByteDance employees can also register, but they are not allowed to participate in the awards). Interested individuals are welcome to participate as a team. Each team can have one or more members (up to 4). Each individual can only be part of one team. The organizers reserve the right to disqualify any participant for misleading information or inappropriate behavior during the challenge.

Participants can register for this grand challenge by either filling out the registration formexternal link or sending their registration information to the organizer via email. Once we receive your information, we will confirm your registration. Please note that we only accept submissions from registered teams. After registration, please download and sign up with Lark and join the topic groupexternal link for all future updates of the challenge and Q&A.

For more detailed information on the challenge task, registration and submission guidelines, please refer to the challenge GitHub repositoryexternal link.


The top three teams will be eligible to have their technical papers included in MMSys proceedings providing the paper passes the quality review before the camera-ready deadline and the author pays the full conference registration. Meanwhile, all winning teams will receive a cash prize sponsored by ByteDance, as presented in the table below. Additionally, all teams with final submissions shall receive gifts from ByteDance.

First Prize Second Prize Third Prize
$4000 $2500 $1500

Challenge Organizers