Proposal

Project Proposal

The Problem

During the peak of the Covid pandemic, online learning became the norm for students across the globe. Teachers would use applications like Zoom and share their screens with the class to give their lectures. However, quite frequently, the professor's screen becomes unreadable for periods of time for some students due to a poor internet connection. This frustrates students who start to struggle to follow what's happening in class. The project aims to look at ways to minimize these times of poor video quality by comparing commonly used and new to the market video codecs.

Previous Works

The work of Katsenu et. al. compares the compression efficiency of 3 video codecs: HEVC-HM, AV1 and VTM, with HEVC used as the benchmark. Compared to HEVC-HM, VTM showed significant bit rate saving than AV1 in both UHD and HD test content. When assessed subjectively using anonymous ratings from participants, AV1 and VTM both showed their edges in different video sequences. Overall, VTM performed better in sequences with more less dynamic content, which is similar to the educational content in traditional classrooms. However, this paper assesses video sequences in slightly unrealistic situations if a low-bandwidth network is taken into consideration: the test content were only in either UHD or HD, whereas given the constraint of the network, content with lower resolution should be studied.


Nguyen et. al. objectively evaluates VVenC, x265 and aomenc AV1 video encoders based on their compression efficiency and encoding time. The HM implementation of HEVC was used as a reference, with VVenC achieved the best compression efficiency and encoding time, followed by AV1 but still better than HEVC. Only x265 performed worse than HEVC even with longer-than-permitted encoding time. However, just like the work of Katsenu et. al., even though VVenC seems like a good candidate to use in low-bandwidth networks, the test content were in UHD and HD.


Barman et. al. compares H.264/MPEG-AVC, H.265/MPEG-HEVC and VP9 Codec in live video streaming for games. In terms of compression efficiency, H.265/MPEG-HEVC fared the best but lost to H.264/MPEGAVC in terms of encoding speed. VP9 performed worse than the other 2 in encoding speed, with compression efficiency varied based on different content. However, live game streaming is a very different environment than online classrooms and care must be taken when evaluating these codecs in remote-teaching context.

The Approach

The plan is to look at the main codecs used on the market such as H.264 or V9 and compare them to newer codecs such as AV1 and JEM. The main differentiating factor in the project is the emphasis on minimally acceptable quality for students to be able to follow. For online lectures, 4K quality with 60fps is overkill, so the emphasis will be placed on lower quality ranges from 480p-1080p at an fps of 15-30fps. Furthermore, since most professors usually teach online by power-point and face cam, the dataset that will be tested will be different from other studies conducted which may lead to different results.

Deliverables

  • Oct. 21: Background research is done on codecs

  • Nov. 4: Objective comparisons completed

  • Nov. 11: Midterm presentation finished

  • Nov.25: Subjective comparisons completed

  • Dec. 2: Final Presentation finished

  • Dec. 9 Final report completed