Enhancing few-shot object detection through pseudo-label mining
Few-shot object detection involves adapting an existing detector to a set of unseen categories with few annotated examples. This data limitation makes these methods to underperform those trained on large labeled datasets. In many scenarios, there is a high amount of unlabeled data that is never expl...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universidad de Santiago de Compostela (USC) |
| Repositorio: | Minerva. Repositorio Institucional de la Universidad de Santiago de Compostela |
| Idioma: | inglés |
| OAI Identifier: | oai:minerva.usc.gal:10347/39982 |
| Acceso en línea: | https://hdl.handle.net/10347/39982 |
| Access Level: | acceso abierto |
| Palabra clave: | Few-shot Object detection Few-shot learning Pseudo-label mining Pseudo-labeling 120304 Inteligencia artificial |
| id |
ES_49f1fc36b28d45fddba8a2a5cb81ee13 |
|---|---|
| oai_identifier_str |
oai:minerva.usc.gal:10347/39982 |
| network_acronym_str |
ES |
| network_name_str |
España |
| spelling |
Enhancing few-shot object detection through pseudo-label mining García Fernández, Pablo Cores Costa, Daniel Mucientes Molina, Manuel Few-shot Object detection Few-shot learning Pseudo-label mining Pseudo-labeling 120304 Inteligencia artificial Few-shot object detection involves adapting an existing detector to a set of unseen categories with few annotated examples. This data limitation makes these methods to underperform those trained on large labeled datasets. In many scenarios, there is a high amount of unlabeled data that is never exploited. Thus, we propose to exPAND the initial novel set by mining pseudo-labels. From a raw set of detections, xPAND obtains reliable pseudo-labels suitable for training any detector. To this end, we propose two new modules: Class and Box confirmation. Class Confirmation aims to remove misclassified pseudo-labels by comparing candidates with expected class prototypes. Box Confirmation estimates IoU to discard inadequately framed objects. Experimental results demonstrate that xPAND enhances the performance of multiple detectors up to +5.9 nAP and +16.4 nAP50 points for MS-COCO and PASCAL VOC, respectively, establishing a new state of the art. Code: https://github.com/PAGF188/xPAND Elsevier https://hdl.handle.net/10347/39982 |
| title |
Enhancing few-shot object detection through pseudo-label mining |
| spellingShingle |
Enhancing few-shot object detection through pseudo-label mining García Fernández, Pablo Few-shot Object detection Few-shot learning Pseudo-label mining Pseudo-labeling 120304 Inteligencia artificial |
| title_short |
Enhancing few-shot object detection through pseudo-label mining |
| title_full |
Enhancing few-shot object detection through pseudo-label mining |
| title_fullStr |
Enhancing few-shot object detection through pseudo-label mining |
| title_full_unstemmed |
Enhancing few-shot object detection through pseudo-label mining |
| title_sort |
Enhancing few-shot object detection through pseudo-label mining |
| author |
García Fernández, Pablo |
| author_facet |
García Fernández, Pablo Cores Costa, Daniel Mucientes Molina, Manuel |
| author_role |
author |
| author2 |
Cores Costa, Daniel Mucientes Molina, Manuel |
| author2_role |
author author |
| topic |
Few-shot Object detection Few-shot learning Pseudo-label mining Pseudo-labeling 120304 Inteligencia artificial |
| topic_facet |
Few-shot Object detection Few-shot learning Pseudo-label mining Pseudo-labeling 120304 Inteligencia artificial |
| description |
Few-shot object detection involves adapting an existing detector to a set of unseen categories with few annotated examples. This data limitation makes these methods to underperform those trained on large labeled datasets. In many scenarios, there is a high amount of unlabeled data that is never exploited. Thus, we propose to exPAND the initial novel set by mining pseudo-labels. From a raw set of detections, xPAND obtains reliable pseudo-labels suitable for training any detector. To this end, we propose two new modules: Class and Box confirmation. Class Confirmation aims to remove misclassified pseudo-labels by comparing candidates with expected class prototypes. Box Confirmation estimates IoU to discard inadequately framed objects. Experimental results demonstrate that xPAND enhances the performance of multiple detectors up to +5.9 nAP and +16.4 nAP50 points for MS-COCO and PASCAL VOC, respectively, establishing a new state of the art. Code: https://github.com/PAGF188/xPAND |
| publishDate |
2025 |
| format |
article |
| url |
https://hdl.handle.net/10347/39982 |
| language |
eng |
| eu_rights_str_mv |
openAccess |
| publisher |
Elsevier |
| institution |
Universidad de Santiago de Compostela (USC) |
| collection |
Minerva. Repositorio Institucional de la Universidad de Santiago de Compostela |
| reponame_str |
Minerva. Repositorio Institucional de la Universidad de Santiago de Compostela |
| instname_str |
Universidad de Santiago de Compostela (USC) |
| _version_ |
1878434239149506560 |
| publishDateSort |
2025 |
| author_browse |
Cores Costa, Daniel García Fernández, Pablo Mucientes Molina, Manuel |
| publisherStr |
Elsevier |
| score |
6,9303427 |