ভিডিও বোঝার কুইজ

আপনি কুইজ নেওয়ার সাথে সাথে উত্তর এবং ব্যাখ্যাগুলি উপস্থিত হয়।

1. A system says an event occurred somewhere in a clip. What additional output does temporal localization require?
   • Only a longer category description
   • Approximate start and end times for the event
   • Only a frame-level object name
   • Only a video file checksum
2. A video is sampled every five seconds. A brief event occurs only from 3.1 to 3.4 seconds. What follows?
   • The event must appear at the five-second frame
   • The clip cannot contain audio
   • The model knows the event from a neighboring frame
   • The sampled frames can miss the event entirely
3. Why does “not seen in sampled frames” fail to prove an event never happened?
   • Unobserved intervals may contain the event
   • Absence in samples is a full-video proof
   • Every frame has identical content
   • Sampling adds missing audio evidence
4. A system identifies a cup and a table but reverses which action came first. Which ability failed?
   • Object recognition only
   • Video compression quality
   • Temporal ordering of events
   • The presence of a cup
5. How can an audio track help a video analysis while also introducing uncertainty?
   • Audio always proves visual intention
   • Audio eliminates the need to inspect frames
   • A transcript is a guaranteed record of every sound
   • It supplies sound context, but transcripts can omit or misattribute speech
6. A model describes a person’s intention from one observed action. What should a reviewer distinguish?
   • Frame rate from file format
   • Visible evidence from inferred mental state
   • Video duration from timestamp
   • Visual objects from subtitles only
7. Which material helps verify a video-summary claim?
   • Only a fluent summary sentence
   • Only the model’s confidence adjective
   • The original clip and linked time range
   • Only an unrelated thumbnail
8. Why test long videos and camera cuts rather than only short demos?
   • They can introduce errors absent from curated short clips
   • Long clips always make models more accurate
   • Cuts supply guaranteed event boundaries
   • Demos are identical to real capture