# Cosine meets Softmax: A tough-to-beat baseline for visual grounding

arXiv: https://arxiv.org/abs/2009.06066 · HTML: https://mlsanity.com/paper/2009.06066/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 13 Sep 2020.

## Built on by (most cited first, 10 of 10)

- [Referring Multi-Object Tracking](https://mlsanity.com/paper/2303.03366/) (2023)
- [GPT-4 Enhanced Multimodal Grounding for Autonomous Driving: Leveraging Cross-Modal Attention with Large Language Models](https://mlsanity.com/paper/2312.03543/) (2023)
- [Ground then Navigate: Language-guided Navigation in Dynamic Scenes](https://mlsanity.com/paper/2209.11972/) (2022)
- [Giving Commands to a Self-Driving Car: How to Deal with Uncertain Situations?](https://mlsanity.com/paper/2106.04232/) (2021)
- [Grounding Linguistic Commands to Navigable Regions](https://mlsanity.com/paper/2112.13031/) (2021)
- [Talk to Parallel LiDARs: A Human-LiDAR Interaction Method Based on 3D Visual Grounding](https://mlsanity.com/paper/2405.15274/) (2024)
- [Predicting Physical World Destinations for Commands Given to Self-Driving Cars](https://mlsanity.com/paper/2112.05419/) (2021)
- [Think Before You Drive: World Model-Inspired Multimodal Grounding for Autonomous Vehicles](https://mlsanity.com/paper/2512.03454/) (2025)
- [E3AD: An Emotion-Aware Vision-Language-Action Model for Human-Centric End-to-End Autonomous Driving](https://mlsanity.com/paper/2512.04733/) (2025)
- [Estimation of Appearance and Occupancy Information in Birds Eye View from Surround Monocular Images](https://mlsanity.com/paper/2211.04557/) (2022)

## Abstract (arXiv)

In this paper, we present a simple baseline for visual grounding for autonomous driving which outperforms the state of the art methods, while retaining minimal design choices. Our framework minimizes the cross-entropy loss over the cosine distance between multiple image ROI features with a text embedding (representing the give sentence/phrase). We use pre-trained networks for obtaining the initial embeddings and learn a transformation layer on top of the text embedding. We perform experiments on the Talk2Car dataset and achieve 68.7% AP50 accuracy, improving upon the previous state of the art by 8.6%. Our investigation suggests reconsideration towards more approaches employing sophisticated attention mechanisms or multi-stage reasoning or complex metric learning loss functions by showing promise in simpler alternatives.