HPatches - Benchmark for local feature evaluation
The HPatches dataset is a set of image sequences under different camera view and illumination conditions for evaluating interest point detectors and local descriptors. The performance is reported with the mean average precision (mAP) on three different tasks: Patch Verification, Image Matching, and Patch Retrieval.
- Patch Verification measures the ability of a descriptor to classify whether two patches are extracted from the same measurement (pairwise classification)
- Image Matching tests to what extent a descriptor can correctly identify correspondences in two images (matching of one against many with hard negatives )
- Patch Retrieval tests how well a descriptor can match a query patch to a large pool of patches extracted from many images, including many distractors (one against many with moderately difficult negatives)
For more information on the methods and the evaluation protocols please see the HPatches GitHub and the paper (PDF):
HPatches: A benchmark and evaluation of handcrafted and learned local descriptors, Vassileios Balntas, Karel Lenc, Andrea Vedaldi and Krystian Mikolajczyk, IEEE TPAMI, 2019, (view PDF)
BreakingNews - Benchmark for article annotation by image and text processing
The dataset contains a variety of news-related information including: the text of the article, captions, related images, part-of-speech tagging, GPS coordinates, semantic topics list or results of sentiment analysis, for about 100K news articles. The figure shows two sample images. All this volume of heterogeneous data makes BreakingNews an appropriate benchmark for several tasks exploring the relation between text and images.
For more information on the methods and the evaluation protocols please see the paper (PDF):
BreakingNews: Article annotation by image and text processing, Arnau Ramisa, Fei Yan, Francesc Moreno-Noguer, Krystian Mikolajczyk, IEEE TPAMI, 2018 (view PDF)