Summary of Vln-video: Utilizing Driving Videos For Outdoor Vision-and-language Navigation, by Jialu Li et al.
VLN-Video: Utilizing Driving Videos for Outdoor Vision-and-Language Navigation
by Jialu Li, Aishwarya Padmakumar, Gaurav Sukhatme, Mohit Bansal
First submitted to arxiv on: 5 Feb 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes VLN-Video, an innovative approach to improve Outdoor Vision-and-Language Navigation (VLN) performance. Existing methods are limited by insufficient environmental diversity and training data. To address this, VLN-Video utilizes driving videos in multiple U.S. cities, augmented with automatically generated navigation instructions and actions. The method combines classical and deep learning techniques, using template infilling for grounded instruction generation and image rotation similarity-based action prediction. Pre-training is performed on the Touchdown dataset and a video-augmented dataset created from driving videos, using three proxy tasks: masked language modeling, instruction and trajectory matching, and next action prediction. The learned instruction representation is adapted to a state-of-the-art navigator for fine-tuning on the Touchdown dataset. Results show that VLN-Video significantly outperforms previous models by 2.1% in task completion rate, achieving a new state-of-the-art on the Touchdown dataset. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about helping computers navigate outdoor environments using natural language instructions. Current methods don’t work well because they lack variety in their training data and environments. To solve this problem, the researchers created VLN-Video, which uses videos of cars driving around cities to generate more diverse navigation instructions and actions. The method combines old and new techniques to make computers better at understanding instructions and making decisions. The results show that VLN-Video is significantly better than previous methods at completing tasks. |
Keywords
» Artificial intelligence » Deep learning » Fine tuning
