Skip to content
Other

EVA-02 Neural Video Retrieval

поиск кадров по позе, ракурсу и одежде

2 min readupdated

Этот чекпоинт решает точный поиск видеокадров по текстовому запросу: важны не только объекты, но и поза тела, ракурс камеры и детали одежды. Он основан на EVA-02-CLIP-L-14-336: vision-энкодер заморожен, а текстовый энкодер и logit_scale дообучены по схеме LiT (обучение только текстовой части). Обучение шло на curated web videos и иллюстрациях Danbooru2023, поэтому модель поддерживает и cross-domain поиск «2D-иллюстрация → реальный кадр». Пригодится для видеоархивов, поисковиков по кадрам и исследовательских retrieval-пайплайнов.

Ключевые особенности

  • Vision-энкодер EVA-02 заморожен (304.43M параметров), дообучены текстовый энкодер и logit_scale (123.65M).
  • Fine-grained retrieval: позы тела, ракурсы камеры, детали гардероба.
  • Поддерживает 2D-to-real cross-domain lookup: поиск реальных кадров по иллюстрациям.
  • MIL-контрастивное обучение InfoNCE на contact sheet 3×3 keyframes, с балансировкой поз и динамическими промптами.
  • На val (n=1,200) Recall@1 вырос с 12.4% до 20.67%, Recall@5 — с 28.1% до 37.83%.

Что можно сделать

  • Собрать текстовый поиск по видеоархиву: запросы про позу, ракурс и одежду, а не только про объекты.
  • Искать реальные кадры по 2D-иллюстрациям или референсам — cross-domain сценарий из карточки.
  • Подключить чекпоинт через open_clip к своему keyframe-retrieval пайплайну.
  • Запустить интерактивный Web Viewer из репозитория, положив best.pt в scrape/siglip_runs/eva02_curated_v2/best.pt.

Пример

import torch, open_clip
from huggingface_hub import hf_hub_download
ckpt = torch.load(hf_hub_download('wheatley1901/eva02-neural-video-retrieval', filename='eva02_curated_v2.pt'), map_location='cpu', weights_only=True)
model, _, preprocess = open_clip.create_model_and_transforms('hf-hub:timm/eva02_large_patch14_clip_336.merged2b_s6b_b61k')
model.load_state_dict(ckpt.get('state_dict', ckpt), strict=True)
model = model.half().cuda().eval()
ПодробнееМетрики на val split (n=1,200), epoch 1: Recall@1 — 20.67% против 12.4% у stock EVA-02; Recall@5 — 37.83% против 28.1%.
Обучение: 147,412 строк curated web videos и 4,652 строк Danbooru2023 illustrations.
Заморожены 304.43M vision-весов; обучены 123.65M параметров текстового энкодера и logit_scale.
Требования к VRAM в карточке не указаны; для Web Viewer нужен файл best.pt в scrape/siglip_runs/eva02_curated_v2/.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.6 GB · ready files2.3 GB
  • FP8 / INT8recommendedweights 0.8 GB1.4 GB
  • 4-bitweights 0.4 GB1.0 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # wheatley1901/eva02-neural-video-retrieval
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download wheatley1901/eva02-neural-video-retrieval --local-dir ~/models/eva02-neural-video-retrieval
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of eva02_large_patch14_clip_336.merged2b_s6b_b61k (quants, finetunes, LoRAs) will show up under the bell.

Fine-tunes 1All on Hugging Face

About the model

wheatley1901/eva02-neural-video-retrieval
Author
wheatley1901
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments