Hierarchical Feature Attention Learning Network for Detecting Object and Discriminative Parts in Fine-Grained Visual Classification
Bibliographic record
Abstract
This paper proposes a novel hierarchical feature attention learning network for improved fine-grained visual classification (FGVC). Existing fine-grained classification methods rely heavily on attention mechanisms to differentiate minute details of similar objects. These mechanisms often assume that critical locations have a similar scale and are uniquely localizable, which is not always accurate. For instance, the size of a bird may vary across images, and the color of its beak might only be significant for species identification when its wing and tail colors are specific. This paper addresses this limitation by proposing a so-called hierarchical feature attention learning network, which initially focuses on the target object within the image, followed by multi-headed attention to identify key discriminative locations (patches). Especially, we develop a novel hierarchical attention approach that appropriately reduces misleading attentions by considering the object’s size for capturing correct attention parts. In addition, the proposed multi-headed attention allows for examining more complementary attention parts to identify the most discriminative features. Further, our framework is implemented as an architectural constraint, eliminating the need for object or part-level annotations in a weakly supervised detection manner. We conducted extensive and comparative experiments on three benchmark datasets: NABirds, CUB-200, and Oxford 102 Flower. The results demonstrate that our proposed hierarchical attention approach provides a robust and efficient solution for improved FGVC. Specifically, our method achieved a top-1 accuracy increase of approximately 93.0%, 92.7%, and 99.4% on the CUB-200-2011, NABirds, and Oxford 102 Flower benchmarks, respectively.
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 | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".