Open Vocabulary Multi-Label Video Classification

TL;DR

This paper introduces a novel method for open vocabulary multi-label video classification by adapting pre-trained vision-language models with semantic guidance from large language models and temporal modeling, enabling recognition of multiple actions and entities in videos.

Contribution

It presents an end-to-end trainable architecture that prompts LLMs for soft attributes and integrates temporal modeling into CLIP for improved multi-label video understanding.

Findings

Effective recognition of multiple actions and objects in videos.

Superior performance on benchmark datasets.

Enhanced open vocabulary classification accuracy.

Abstract

Pre-trained vision-language models (VLMs) have enabled significant progress in open vocabulary computer vision tasks such as image classification, object detection and image segmentation. Some recent works have focused on extending VLMs to open vocabulary single label action classification in videos. However, previous methods fall short in holistic video understanding which requires the ability to simultaneously recognize multiple actions and entities e.g., objects in the video in an open vocabulary setting. We formulate this problem as open vocabulary multilabel video classification and propose a method to adapt a pre-trained VLM such as CLIP to solve this task. We leverage large language models (LLMs) to provide semantic guidance to the VLM about class labels to improve its open vocabulary performance with two key contributions. First, we propose an end-to-end trainable architecture that learns to prompt an LLM to generate soft attributes for the CLIP text-encoder to enable it to recognize novel classes. Second, we integrate a temporal modeling module into CLIP's vision encoder to effectively model the spatio-temporal dynamics of video concepts as well as propose a novel regularized finetuning technique to ensure strong open vocabulary classification performance in the video domain. Our extensive experimentation showcases the efficacy of our approach on multiple benchmark datasets.