# Learning Dense Visual Descriptors using Image Augmentations for Robot Manipulation Tasks

arXiv: https://arxiv.org/abs/2209.05213 · HTML: https://mlsanity.com/paper/2209.05213/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 12 Sep 2022.

## Built on by (most cited first, 7 of 7)

- [Decomposing the Generalization Gap in Imitation Learning for Visual Robotic Manipulation](https://mlsanity.com/paper/2307.03659/) (2023)
- [One-Shot Imitation Learning with Invariance Matching for Robotic Manipulation](https://mlsanity.com/paper/2405.13178/) (2024)
- [The Treachery of Images: Bayesian Scene Keypoints for Deep Policy Learning in Robotic Manipulation](https://mlsanity.com/paper/2305.04718/) (2023)
- [Doduo: Learning Dense Visual Correspondence from Unsupervised Semantic-Aware Flow](https://mlsanity.com/paper/2309.15110/) (2023)
- [CLAMP: Contrastive Learning for 3D Multi-View Action-Conditioned Robotic Manipulation Pretraining](https://mlsanity.com/paper/2602.00937/) (2026)
- [Cycle-Correspondence Loss: Learning Dense View-Invariant Visual Features from Unlabeled and Unordered RGB Images](https://mlsanity.com/paper/2406.12441/) (2024)
- [ManiLong-Shot: Interaction-Aware One-Shot Imitation Learning for Long-Horizon Manipulation](https://mlsanity.com/paper/2512.16302/) (2025)

## Abstract (arXiv)

We propose a self-supervised training approach for learning view-invariant dense visual descriptors using image augmentations. Unlike existing works, which often require complex datasets, such as registered RGBD sequences, we train on an unordered set of RGB images. This allows for learning from a single camera view, e.g., in an existing robotic cell with a fix-mounted camera. We create synthetic views and dense pixel correspondences using data augmentations. We find our descriptors are competitive to the existing methods, despite the simpler data recording and setup requirements. We show that training on synthetic correspondences provides descriptor consistency across a broad range of camera views. We compare against training with geometric correspondence from multiple views and provide ablation studies. We also show a robotic bin-picking experiment using descriptors learned from a fix-mounted camera for defining grasp preferences.