English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Few-shot object detection (FSOD) benchmarks have advanced techniques fordetecting new categories with limited annotations. Existing benchmarksrepurpose well-established datasets like COCO by partitioning categories intobase and novel classes for pre-training and fine-tuning respectively. However,these benchmarks do not reflect how FSOD is deployed in practice. Rather thanonly pre-training on a small number of base categories, we argue that it ismore practical to fine-tune a foundation model (e.g., a vision-language model(VLM) pre-trained on web-scale data) for a target domain. Surprisingly, we findthat zero-shot inference from VLMs like GroundingDINO significantly outperformsthe state-of-the-art (48.3 vs. 33.1 AP) on COCO. However, such zero-shot modelscan still be misaligned to target concepts of interest. For example, trailerson the web may be different from trailers in the context of autonomousvehicles. In this work, we propose Foundational FSOD, a new benchmark protocolthat evaluates detectors pre-trained on any external datasets and fine-tuned onK-shots per target class. Further, we note that current FSOD benchmarks areactually federated datasets containing exhaustive annotations for each categoryon a subset of the data. We leverage this insight to propose simple strategiesfor fine-tuning VLMs with federated losses. We demonstrate the effectiveness ofour approach on LVIS and nuImages, improving over prior work by 5.9 AP.

Revisiting Few-Shot Object Detection with Vision-Language Models