This repository hosts the artifacts (config, tokenizer, weights) for the GROVE model, as presented in the paper Large-scale Pre-training for Grounded Video Caption Generation.

Project Website: https://ekazakos.github.io/grounded_video_caption_generation/ Codebase: The inference code is provided in the grove-transformers package — a slimmer version of the full codebase at https://github.com/ekazakos/grove/, designed specifically for quick inference with GROVE.

We propose a novel approach for captioning and object grounding in video, where the objects in the caption are grounded in the video via temporally dense bounding boxes. We introduce the following contributions. First, we present a large-scale automatic annotation method that aggregates frame-level captions grounded with bounding boxes into temporally dense and consistent annotations. We apply this approach on the HowTo100M dataset to construct a large-scale pre-training dataset, named HowToGround1M. We also introduce a Grounded Video Caption Generation model, dubbed GROVE, and pre-train the model on HowToGround1M. Second, we introduce iGround--a dataset of 3513 videos with manually annotated captions and dense spatio-temporally grounded bounding boxes. This allows us to measure progress on this challenging problem, as well as to fine-tune our model on this small-scale but high-quality data. Third, we demonstrate that our approach achieves state-of-the-art results on the proposed iGround dataset, as well as on the VidSTG, ActivityNet-Entities, GroundingYouTube, and YouCook-Interactions datasets. Our ablations demonstrate the importance of pre-training on our automatically annotated HowToGround1M dataset followed by fine-tuning on the manually annotated iGround dataset and validate the key technical contributions of our model. The dataset and code are available at this https URL .

Install the inference package:

If you've already cloned the main repo from https://github.com/ekazakos/grove/, then run:

Alternatively, run:

Also, install mmcv, mmdetection and SAM2 as shown here.

If you don’t have precomputed token embeddings for GROVE's vocabulary or video metadata, just pass the video path.
GROVE will compute everything internally.

If you have precomputed token embeddings for GROVE's vocabulary and video metadata (e.g. from datasets like HowToGround1M or iGround), you can pass them directly for faster inference and precise trimming.