VIVA lab - University of Ottawa at TRECVID 2009 Content Based Copy Detection
Bibliographic record
Abstract
Briefly, what approach or combination of approaches did you test in each of your submitted runs? † VIVAlab-uOttawa.v The video search approach is based on a method that finds keypoints in each video frame, and then uses a descriptor of keypoint counts in 16 (4 by 4) equally sized regions of the images. † VIVAlab-uOttawa.a The audio-only copy detection search scheme was based on the computation of a coherence function, using intermediate features in the ITU-R BS.1387 Perceptual Evaluation of Audio Quality (PEAQ) standard. † VIVAlab-uOttawa.m The combined audio-video used the fact that audio was detected in a database segment simply to boost the video scores. 2. What if any significant differences (in terms of what measures) did you find among the runs? The video only approach performs well in terms of the mean F1 measure, which is a consequence of the construction of the algorithm. Indeed, because a frame is represented with only 16 bytes, it was possible to compare each frame of the query with each frame of the video database. As a result, our algorithm is one of the best in terms of accuracy. The balanced profile performs at almost exactly the median in terms of NDCR, while the NOFA profile is slightly worse than the median performance. The audio only approach is slightly worse than the median in terms of NDCR. The F1 results are not significant in this case because of the very low number of positive detection that were submitted. The audio detection method has a very low false positive detection rate but, unfortunately a very high false negative too. The combined results were lower than what we expected; we still have to analyze the root cause of this performance. 3. Based on the results, can you estimate the relative contribution of each component of your system/approach to its effectiveness? Currently the video analysis seems to be doing most of the work, but we do not believe that our combination of the audio and video search methods is optimal. 4. Overall, what did you learn about runs/approaches and the research question(s) that motivated them? With so many different possible solutions to copy detection deciding which is the best is a function of the goals. Our goal is to have an approach that provides good performance, is easy to parallelize, works quickly in search mode, and has low storage requirements per video frame of the database. We believe that we have achieved this goals, but we have not yet found a satisfactory way of combining the audio video results.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.014 |
| Meta-epidemiology (narrow) | 0.007 | 0.002 |
| Meta-epidemiology (broad) | 0.005 | 0.003 |
| Bibliometrics | 0.006 | 0.005 |
| Science and technology studies | 0.003 | 0.001 |
| Scholarly communication | 0.007 | 0.007 |
| Open science | 0.008 | 0.004 |
| Research integrity | 0.004 | 0.005 |
| Insufficient payload (model declined to judge) | 0.063 | 0.106 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".