[Models](/models)/[Segmentation](/models?category=segmentation)/CIDAS/clipseg-rd64-refined

HFSegmentationApache 2.0

# clipseg-rd64-refined

by CIDAS

Text-prompted image segmentation for queryable masks and region crops

1.7Mdl/month

0.2Bparams

[HuggingFace](https://huggingface.co/CIDAS/clipseg-rd64-refined)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`CIDAS/clipseg-rd64-refined`

## Deploy clipseg-rd64-refined

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

CLIPSeg RD64 Refined is a CLIP-conditioned segmentation model that produces a mask from an image plus a natural language prompt. Instead of requiring a fixed class label set, it lets a pipeline ask for regions like "red logo," "person holding a box," or "damaged corner" and turn those regions into indexed evidence.

On Mixpeek, CLIPSeg is useful before region embedding or visual QA. The segmenter isolates the queried foreground, Mixpeek stores the mask geometry and crop lineage, and an agent can search or inspect the precise region instead of the whole frame.

## Architecture

CLIPSeg combines a CLIP visual-text backbone with a lightweight decoder for dense prediction. The RD64 refined checkpoint is optimized for image segmentation with natural language prompts and outputs pixel masks aligned to the input image.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so clipseg-rd64-refined runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace\_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "asset-00412",
          // Boxes, masks, depth maps and anomaly scores are structured
          // results, not vectors. They go in payload and are reachable
          // through pre\_filters on a retriever, not through similarity.
          payload: {
            detections: modelOutput,
            source\_key: "archive/2026/asset-00412",
          },
        },
      \],
    }),
  },
);

// No managed alternative for an open label set. Two extractors do emit a
// bbox, for the one thing each detects: document\_graph\_extractor@v1 per
// layout block, face\_identity\_extractor@v1 per face. Nothing ships that
// returns masks, depth maps or anomaly scores.

## Capabilities

*   Text-guided segmentation without a closed class list
*   Foreground mask generation for natural images and video keyframes
*   Region crop extraction before visual embedding
*   Spatial metadata for evidence citations

## Use Cases on Mixpeek

Find frames where a specific described object appears

Crop queried regions before CLIP, SigLIP, or Nomic vision embedding

Build agent tools that inspect only the visual region relevant to a question

Filter product, ad, or screenshot libraries by prompted visual regions

## Benchmarks

Dataset

Metric

Score

Source

PhraseCut

Segmentation

\-

CLIPSeg paper

RefCOCO

Referring segmentation

\-

CLIPSeg paper

## Performance

Input SizeImage plus text prompt

GPU Latency~20-40ms / image on A100

GPU ThroughputBatch dependent

GPU Memory~1 GB

Run on selected frames or first-stage candidates when prompt count is high

## Common Pipeline Companions

[

nomic-ai/nomic-embed-vision-v1.5

Embed foreground crops for region search

](/model/nomic-ai/nomic-embed-vision-v1.5)[

Qwen/Qwen3-VL-Reranker-2B

Rerank cropped visual evidence with the query

](/model/Qwen/Qwen3-VL-Reranker-2B)[

ZhengPeng7/BiRefNet

Fallback foreground segmentation when no text prompt is needed

](/model/ZhengPeng7/BiRefNet)

## Explore on Mixpeek

[

More Segmentation models

Compare alternatives in this category

](/models?category=segmentation)[

Best Computer Vision APIs

Hand-picked tools & platforms compared

](/curated-lists/best-computer-vision-apis)[

Mask-Aware Retrieval

Deep-dive technical guide

](/guides/mask-aware-retrieval-for-ai-agents)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

FrameworkHF

OrganizationCIDAS

FeatureSegmentation

Outputmask + label

Modalitiesvideo, image

RetrieverMask Filter

Parameters0.2B

LicenseApache 2.0

Downloads/mo1.7M

[

### Research Paper

Image Segmentation Using Text and Image Prompts

arxiv.org](https://arxiv.org/abs/2112.10003)

### Build a pipeline with clipseg-rd64-refined

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

### Alternative Models

[

facebook/sam-vit-huge

Segmentation

](/model/facebook/sam-vit-huge)[

facebook/sam2.1-hiera-large

Segmentation

](/model/facebook/sam2.1-hiera-large)[

facebook/sam3

Segmentation

](/model/facebook/sam3)[

facebook/sam3.1

Segmentation

](/model/facebook/sam3.1)