Totally-Looks-Like: A Dataset and Benchmark of Semantic Image Similarity
Bibliographic record
Abstract
Human perception of images goes far beyond objects, shapes, textures and contours. Viewing a scene often elicits recollection of other scenes whose global properties or relations resemble the currently observed one. This relies on a rich representation in image space in the brain, entailing scene structure and semantics, as well as a mechanism to use the representation of an observed scene to recollect similar ones from the profusion of those stored in memory. The recent explosion in the performance and applicability of deep-learning models in all fields of computer vision, including image retrieval and comparison, can tempt one to conclude that the representational power of such methods approaches that of humans. We aim to explore this by testing how deep neural networks fare on the challenge of similarity judgement between pairs of images from a new dataset, dubbed "Totally-Looks-Like". It is based on images from a website in popular media, which hosts pairs of images deemed by users to appear similar to each other, though they often share little common appearance, if judging by low-level visual features. These include pairs of images out of (but not limited to) objects, scenes, patterns, animals, and faces across various modalities (sketch, cartoon, natural images). The website also includes user ratings, showing the level of agreement with the proposed resemblances. The dataset is very diverse and implicitly represents many aspects of human perception of image similarity. We evaluate the performance of several state-of-the-art models on this dataset, comparing their performance with human similarity judgements. The comparison not only forms a benchmark for other similar evaluations, but also reveals specific weaknesses in the strongest of the current systems that point the way for future research. Meeting abstract presented at VSS 2018
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 | 0.002 | 0.007 |
| Meta-epidemiology (narrow) | 0.004 | 0.000 |
| Meta-epidemiology (broad) | 0.002 | 0.004 |
| Bibliometrics | 0.006 | 0.005 |
| Science and technology studies | 0.002 | 0.002 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.005 | 0.004 |
| Research integrity | 0.005 | 0.003 |
| Insufficient payload (model declined to judge) | 0.007 | 0.006 |
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".