ELEVATER: A Benchmark and Toolkit for Evaluating Language-Augmented Visual Models

Chunyuan Li, Haotian Liu, Liunian Harold Li, Pengchuan Zhang, Jyoti Aneja, Jianwei Yang, Ping Jin, Houdong Hu, Zicheng Liu, Yong Jae Lee, Jianfeng Gao

Keywords: evaluation platform, task-level transfer, language-image pre-training, image classification, object detection

TL;DR: ELEVATER provides the first public platform and toolkit to evaluate vision foundation models in their large-scale task-level visual transfer in 20 image classification tasks and 35 object detection tasks

Abstract: Learning visual representations from natural language supervision has recently shown great promise in a number of pioneering works. In general, these language-augmented visual models demonstrate strong transferability to a variety of datasets/tasks. However, it remains challenging to evaluate the transferablity of these foundation models due to the lack of easy-to-use toolkits for fair benchmarking. To tackle this, we build ELEVATER (Evaluation of Language-augmented Visual Task-level Transfer), the first benchmark to compare and evaluate pre-trained language-augmented visual models. Several highlights include: (i) Datasets. As downstream evaluation suites, it consists of 20 image classification datasets and 35 object detection datasets, each of which is augmented with external knowledge. (ii) Toolkit. An automatic hyper-parameter tuning toolkit is developed to ensure the fairness in model adaption. To leverage the full power of language-augmented visual models, novel language-aware initialization methods are proposed to significantly improve the adaption performance. (iii) Metrics. A variety of evaluation metrics are used, including sample-efficiency (zero-shot and few-shot) and parameter-efficiency (linear probing and full model fine-tuning). We will publicly release ELEVATER.

Author Statement: Yes

URL: https://computer-vision-in-the-wild.github.io/ELEVATER/

Dataset Url: https://github.com/Computer-Vision-in-the-Wild/DataDownload

License:

Our code is under MIT license.
The data of our benchmark includes two parts: (1) A suite of public datasets. Please follow the original license of each dataset. (2) External knowledge. The knowledge extracted from WordNet/Wiktionary follows their own licenses. For the knowledge extracted from GPT3, we have the approval from OpenAI to use it for this research benchmark to encourage future studies.

Contribution Process Agreement: Yes

In Person Attendance: No