Vision-Language Model for Object Detection and Segmentation: A Review and Evaluation
September 21, 2026 ยท View on GitHub
This is the repository of Vision-Language Model for Object Detection and Segmentation: A Review and Evaluation
The four adaptation paradigms evaluated here. Every detector is run under all four, and they differ only in which components receive gradients: dashed grey is frozen, solid blue is updated, orange is the learnable prompt introduced by the paradigm itself.
Evaluation Radar Results. Each axis is one task setting; the pair beneath an axis label gives the smallest and the largest value any plotted model attains on that axis, divided by 100, and the radius is that value rescaled linearly between the two. Radii are therefore comparable along an axis but not between axes.
Models
Last update on 2026/09/21
VLM Detection Models
First Type
- [CVPR 2022] Grounded Language-Image Pre-training [Paper][Code]
- [CVPR 2022] RegionCLIP: Region-based Language-Image Pretraining [Paper][Code]
- [ECCV 2022] Open Vocabulary Object Detection with Pseudo Bounding-Box Labels [Paper][Code]
- [NeulIPS 2022] DetCLIP: Dictionary-Enriched Visual-Concept Paralleled Pre-training for Open-world Detection [Paper]
- [ECCV 2022] Simple Open-Vocabulary Object Detection with Vision Transformers [Paper][Code]
- [NeurIPS 2023] Scaling Open-Vocabulary Object Detection [Paper][Code]
- [CVPR 2023] DetCLIPv2: Scalable Open-Vocabulary Object Detection Pre-training via Word-Region Alignment [Paper]
- [CVPR 2024] DetCLIPv3: Towards Versatile Generative Open-vocabulary Object Detection [Paper]
- [ECCV 2024] Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection [Paper][Code]
- [CVPR 2024] YOLO-World: Real-Time Open-Vocabulary Object Detection [Paper][Code]
- [arxiv] OV-DINO: Unified Open-Vocabulary Detection with Language-Aware Selective Fusion [Paper][Code]
Second Type
-
[ECCV 2022] Detecting Twenty-thousand Classes using Image-level Supervision [Paper][Code]
-
[ICLR 2023] Learning Object-Language Alignments for Open-Vocabulary Object Detection [Paper][Code]
-
[CVPR 2022] Learning to Prompt for Open-Vocabulary Object Detection with Vision-Language Model [Paper][Code]
-
[ECCV 2022] Open-Vocabulary DETR with Conditional Matching [Paper][Code]
-
[ICLR 2022] Open-vocabulary Object Detection via Vision and Language Knowledge Distillation [Paper][Code]
-
[CVPR 2022] Open-Vocabulary One-Stage Detection with Hierarchical Visual-Language Knowledge Distillation [Paper][Code]
-
[ECCV 2022] PromptDet: Towards Open-vocabulary Detection using Uncurated Images [Paper][Code]
-
[CVPR 2023] Aligning Bag of Regions for Open-Vocabulary Object Detection [Paper][Code]
-
[NeurIPS 2023] CoDet: Co-Occurrence Guided Region-Word Alignment for Open-Vocabulary Object Detection [Paper][Code]
-
[CVPR 2023] CORA: Adapting CLIP for Open-Vocabulary Detection with Region Prompting and Anchor Pre-Matching [Paper][Code]
-
[ICCV 2023] Distilling DETR with Visual-Linguistic Knowledge for Open-Vocabulary Object Detection [Paper][Code]
-
[arxiv] DST-Det: Simple Dynamic Self-Training for Open-Vocabulary Object Detection[Paper][Code]
-
[ICCV 2023] EdaDet: Open-Vocabulary Object Detection Using Early Dense Alignment [Paper]
-
[ICLR 2023] F-VLM: Open-Vocabulary Object Detection upon Frozen Vision and Language Models [Paper][Code]
-
[ICML 2023] Multi-Modal Classifiers for Open-Vocabulary Object Detection [Paper][Code]
-
[CVPR 2023] Object-Aware Distillation Pyramid for Open-Vocabulary Object Detection [Paper][Code]
-
[arxiv] Prompt-Guided Transformers for End-to-End Open-Vocabulary Object Detection [Paper]
-
[CVPR 2023] Region-Aware Pretraining for Open-Vocabulary Object Detection with Vision Transformers [Paper][Code]
-
[CVPR 2024] Taming Self-Training for Open-Vocabulary Object Detection [Paper][Code]
-
[ICLR 2024] CLIPSelf: Vision Transformer Distills Itself for Open-Vocabulary Dense Prediction [Paper][Code]
-
[WACV 2024] LP-OVOD: Open-Vocabulary Object Detection by Linear Probing [Paper][Code]
VLM Segmentation Models
- [ICLR 2022] Language-driven Semantic Segmentation [Paper][code]
- [CVPR 2024] CAT-Seg: Cost Aggregation for Open-Vocabulary Semantic Segmentation [Paper][code]
- [CVPR 2023] Side Adapter Network for Open-Vocabulary Semantic Segmentation [Paper]code
- [ECCV 2022] A Simple Baseline for Open-Vocabulary Semantic Segmentation with Pre-trained Vision-language Model [Paper] [code]
- [ICML 2023] Open-Vocabulary Universal Image Segmentation with MaskCLIP[Paper][code]
- [ICCV 2023] Open-Vocabulary Semantic Segmentation with Decoupled One-Pass Network[Paper][code]
- [NeurIPS 2023] Convolutions Die Hard: Open-Vocabulary Segmentation with Single Frozen Convolutional CLIP[Paper] [code]
- [NeurIPS 2023] Learning Mask-aware CLIP Representations for Zero-Shot Segmentation [Paper] [code]
- [CVPR 2024] SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation[Paper][code]
- [CVPR 2024] Open-Vocabulary Segmentation with Semantic-Assisted Calibration[Paper][code]
- [CVPR 2024] Transferable and Principled Efficiency for Open-Vocabulary Segmentation[Paper] [code]
- [CVPR 2024] Open-Vocabulary Semantic Segmentation with Image Embedding Balancing[Paper][code]
- [CVPR 2022] Decoupling Zero-Shot Semantic Segmentation[Paper][code]
- [CVPR 2023] ZegCLIP: Towards Adapting CLIP for Zero-shot Semantic Segmentation[Paper] [code]
- [CVPR 2023] Primitive Generation and Semantic-related Alignment for Universal Zero-Shot Segmentation[Paper] [code]
- [ICML 2024] Cascade-CLIP: Cascaded Vision-Language Embeddings Alignment for Zero-Shot Semantic Segmentation[Paper] [code]
- [ICML2024] SegCLIP: Patch Aggregation with Learnable Centers for Open-Vocabulary Semantic Segmentation[Paper] [code]
- [CVPR 2023] Learning to Generate Text-grounded Mask for Open-world Semantic Segmentation from Only Image-Text Pairs[Paper] [code]
Datasets
Datasets for Detection
| Dataset | Year | Classes | Training | Testing | Evaluation Metric | Project |
|---|---|---|---|---|---|---|
| COCO 2014 Detection | 2014 | 80 | 83,000 | 41,000 | Box mAP | Project |
| COCO 2017 Detection | 2017 | 80 | 118,287 | 5,000 | Box mAP | Project |
| PASCAL VOC | 2012 | 20 | 5,717 | 5,823 | Box mAP | Project |
| LVIS | 2019 | 1203 | 100,170 | 19,809 | Box mAP | Project |
| ODinW | 2022 | 314 | 132,413 | 20,070 | Box mAP | Project |
| Objects365 | 2019 | 365 | 600,000 | 38,000 | Box mAP | Project |
| Stanford Dogs | 2011 | 120 | 12,000 | 8,580 | Box mAP | Project |
| CUB-200-2011 | 2011 | 200 | 5,994 | 5,794 | Box mAP | Project |
| Cityscapes | 2016 | 8 | 2,975 | 500 | Box mAP | Project |
| Foggy Cityscapes | 2018 | 8 | 2,975 | 500 | Box mAP | Project |
| WaterColor | 2018 | 6 | 1,000 | - | Box mAP | Project |
| Comic | 2018 | 6 | 1,000 | - | Box mAP | Project |
| KITTI | 2012 | 1 | 7,481 | - | Box mAP | Project |
| Sim10K | 2016 | 1 | 10,000 | - | Box mAP | Project |
| VOC-C | 2019 | 20 | 543,115 | 553,185 | Box mAP | Project |
| COCO-C | 2019 | 80 | 11,237,265 | 475,000 | Box mAP | Project |
| Cityscapes-C | 2019 | 8 | 282,625 | 47,500 | Box mAP | Project |
| CrowdHuman | 2018 | 1 | 15,000 | 4,370 | Box mAP | Project |
| OCHuman | 2019 | 1 | - | 2,500 | Box mAP | Project |
| WiderPerson | 2019 | 1 | 7,891 | 1,000 | Box mAP | Project |
Datasets for Segmentation
| Dataset | Year | Classes | Training | Testing | Evaluation Metric | Project |
|---|---|---|---|---|---|---|
| COCO-Stuff | 2018 | 172 | 118k | 20k | mIoU | Project |
| PASCAL VOC 2012 | 2012 | 20 | 1,464 | 1,449 | mIoU | Project |
| PASCAL Content | 2014 | 459 | 4,998 | 5,105 | mIoU | Project |
| Cityscapes | 2016 | 19 | 2,975 | 500 | mIoU | Project |
| ADE20k | 2017 | 150 | 25,574 | 2,000 | mIoU | Project |
| MESS* | 2023 | - | - | - | mIoU | Project |
| PASCAL-Part | 2023 | 116 | 8432 | 851 | mIoU | Project |
| ADE20k-Part-234 | 2023 | 234 | 7348 | 1017 | mIoU | Project |
| PASCAL-5i** | 2015 | 20 | - | - | mIoU, FB-IoU | Project |
| COCO-20i** | 2014 | 80 | - | - | mIoU, FB-IoU | Project |
| FSS-1000 | 2020 | 1000 | 5200 | 2400 | mIoU, FB-IoU | Project |
| OCHuman | 2019 | 1 | - | 2231 | AP, AP50, AP75 | Project |
| CIS | 2023 | 1 | - | 459 | AP, AP50, AP75 | Project |
| COCO-OCC | 2021 | 80 | - | 1005 | AP, AP50, AP75 | Project |
| CamVid | 2008 | 11 | 467 | 233 | mIoU | Project |
| UAVid | 2018 | 9 | 200 | 100 | mIoU | Projects |
| UDD6 | 2018 | 12 | 205 | 45 | mIoU | Project |
*The benchmark includes a wide range of domain-specific datasets.
**The benchmark has different training and testing sets under various settings.
Bibtex
If our work is helpful for your research, please consider citing the following BibTeX entry.
@article{feng2025vision,
title={Vision-Language Model for Object Detection and Segmentation: A Review and Evaluation},
author={Feng, Yongchao and Liu, Yajie and Yang, Shuai and Cai, Wenrui and Zhang, Jinqing and Zhan, Qiqi and Huang, Ziyue and Yan, Hongxi and Wan, Qiao and Liu, Chenguang and others},
journal={arXiv preprint arXiv:2504.09480},
year={2025}
}