Full Transcript
Loading transcript…
Buổi họp 2 - 19/7/2026
Watch with subtitles, summary & AI chat
Add the free Subkun extension — works directly on YouTube.
- Watch
- Subtitles
- Summary
- Ask AI
Computer science students and researchers in AI/computer vision facing project deadlines.
TL;DR
A frustrated advisor reviews a student team's research progress on computer vision (detection, tracking, re-identification) for a video surveillance system. The demo fails, deadlines loom, and the advisor demands a clear pipeline diagram and functional prototype within a week.
Key Takeaways
- The advisor requires the team to draw a system architecture diagram before implementation, because without a diagram the research problem cannot be defined.
- The research team wasted 5-6 weeks with no real progress, only demonstrating a buggy interface and no results for tracking people in video.
- The advisor suggests the team choose one of three problems: detection, tracking, or embedding learning, and decides on embedding learning for the team.
- The team's search interface is criticized because the input field is too short, unsuitable for long queries, while major companies like Google design wide search fields.
- The advisor threatens to fail the team if there is no complete demo by next week, since the mid-term report deadline has already passed.
- The meeting reveals the team's lack of preparation and testing, as the demo video does not show people in bounding boxes, leading to the advisor's frustration.
In This Video
- 00:00Meeting opening
Discussion on choosing a research topic and other groups.
- 01:01Research problem analysis
Advisor selects one of three problems: detection, tracking, or embedding learning.
- 02:12System diagram requirement
Emphasizes the need to draw a diagram to clearly define research steps.
- 04:13Search system demo
Presentation of text search functionality and returned results.
- 07:01Video display error
Bug discovered: people not displayed in tracking video.
- 08:52Next week report plan
Sets report deadline and reassigns tasks.
- 10:44Interface fixes and wrap-up
Requests input field improvement and schedules next meeting.
Questions & Answers
How to choose a research problem?
The advisor requires choosing one of three: detection, tracking, or embedding learning. He suggests embedding learning as new, reusing existing models for detection/tracking.
Why must we draw a research diagram?
The advisor stresses drawing a pipeline to clearly define offline and online steps; without it, research cannot proceed in the right direction.
When is the report deadline?
September 15 is the submission date. The advisor gives two weeks to complete the demo and diagram, otherwise the team will fail.
What is wrong with the demo?
The video does not show people in bounding boxes, tracking is not optimized, and results are incorrect. The advisor demands immediate fixes.
How should the search interface be improved?
The advisor compares it to Google: only one long input field is needed, not many cluttered components. The current input field is too small.
What must be reported next week?
Present the overall diagram again, redraw the pipeline. One person handles the application, the other chooses the research problem (the advisor chooses embedding).
Key Terms
- detection — Detecting objects (human detect) within video frames.
- tracking — Tracking objects across frames, possibly using algorithms like bytrack.
- embedding learning — Learning feature representations of images; the advisor proposes choosing this problem.
Download or copy the punctuated YouTube transcript (Markdown)
Full Transcript
Loading transcript…
Full Transcript (Bilingual)
Loading transcript…