# OpenGVLab/InternVideo

[ECCV2024] Video Foundation Models & Data for Multimodal Understanding

Repository: https://github.com/OpenGVLab/InternVideo
Canonical: https://ross.abutalabs.com/products/internvideo
Language: Python
License: Apache-2.0
License Family: permissive
Topics: foundation-models, video-understanding, vision-transformer, action-recognition, masked-autoencoder, multimodal, open-set-recognition, spatio-temporal-action-localization, temporal-action-localization, video-question-answering, video-retrieval, zero-shot-classification, zero-shot-retrieval, benchmark, contrastive-learning, self-supervised, instruction-tuning, video-data, video-dataset, video-clip
Last push: 2026-07-02T03:20:41+00:00

## Health v2 (maintenance only)
Score: 72/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 90, release rhythm 35, longevity 98
- inputs: {"age_days": 1379, "days_push": 62, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2368, forks 159 (observed 2026-08-28T04:06:41.668440+00:00)

## What it is
InternVideo is a series of open-source video foundation models for multimodal video understanding, spanning generative and discriminative learning, long-context video MLLMs, and video agents. It also releases InternVid, a large-scale video-text dataset with hundreds of millions of pairs for training and evaluation.

## Use cases
- understand and answer questions about videos with a multimodal LLM
- retrieve videos using text queries
- classify human actions in video clips
- localize actions temporally in long videos
- train a video-text model on a large video dataset
- run zero-shot video classification benchmarks
- build an agent that reasons over long video content

## When to choose
- you need state-of-the-art pretrained video understanding models with released checkpoints
- you want a large-scale video-text dataset for pretraining or fine-tuning
- you need video retrieval, QA, or action recognition in one model family
- you are researching multimodal video-language models

## When to avoid
- you only need lightweight real-time video processing without deep models
- you lack GPU resources, as the flagship models are 1B-8B parameters
- you need production video streaming or transcoding rather than understanding
- you need a plug-and-play app rather than research code

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, video-processing, nlp, rag, agent-framework
- domain: artificial-intelligence, deep-learning, computer-vision, large-language-models
- platform: python, cross-platform
- tags: video-foundation-models, multimodal, video-text-dataset, video-retrieval, video-question-answering, action-recognition, zero-shot-classification, contrastive-learning, self-supervised-learning, instruction-tuning, vision-transformer, eccv2024, video, ai-agents, gpu, linux

## Member repositories
- OpenGVLab/InternVideo (main) score 72

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:41.668440+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:35:28.185807+00:00, confidence not recorded.
  - readme: https://github.com/OpenGVLab/InternVideo (fetched 2026-08-28T04:06:41.668440+00:00, sha 3981414a9e0b)
- Data as of 2026-08-30T08:39:29.467469+00:00.