DISK: Learning local features with policy gradient

By M. Tyszkiewicz, Pascal Fua, Eduard Trulls. Neural Information Processing Systems, 2020

600 citations recorded by the source corpus.

Local feature frameworks are difficult to learn in an end-to-end fashion, due to the discreteness inherent to the selection and matching of sparse keypoints. We introduce DISK (DIScrete Keypoints), a novel method that overcomes these obstacles by leveraging principles from Reinforcement Learning (RL), optimizing end-to-end for a high number of correct feature matches. Our simple yet expressive probabilistic model lets us keep the training and inference regimes close, while maintaining good enough convergence properties to reliably train from scratch. Our features can be extracted very densely while remaining discriminative, challenging commonly held assumptions about what constitutes a good keypoint, as showcased in Fig. 1, and deliver state-of-the-art results on three public benchmarks.

Subfields: Computer Vision and Pattern Recognition, Geography, Planning and Development

View Paper: https://www.semanticscholar.org/paper/80bc689b89da77f5b749e5ee17a363d7575ae936

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

CoTracker3: Simpler and Better Point Tracking by Pseudo-Labeling Real Videos (2025, unknown): “5, we have explored alternative point sampling methods, including LightGlue [20], Super-Point [5], and DISK [36].”
A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondence (2023, background): “Deep models [15, 40, 45, 52, 66, 71] learn effective features for dense correspondence that can be robust to photometric and geometric changes such as rotation, scaling, or perspective transformation.”
RoMa: Robust Dense Feature Matching (2023, unknown): “Feature matching has traditionally been approached by key-point detection and description followed by matching the descriptions [4, 14, 33, 39, 41, 53].”
VGGT: Visual Geometry Grounded Transformer (2025, unknown): “In recent years, deep learning has improved many components of the SfM pipeline, with keypoint detection [12, 19, 76, 102] and image matching [7, 46, 60, 66] being two primary areas of focus.”
LoFTR: Detector-Free Local Feature Matching with Transformers (2021, methodology): “Many learning-based local features along this line [32, 11, 25, 28, 47] also adopt the detectorbased design.”
DUSt3R: Geometric 3D Vision Made Easy (2023, unknown): “These improvements encompass advanced feature description [26, 28, 97, 135, 167], more accurate image matching [3, 17, 60, 82, 100, 120, 126, 145], featuremetric refinement [59], and neural bundle adjustment [58, 153].”