CLIPer: Hierarchically Improving Spatial Representation of CLIP for Open-Vocabulary Semantic Segmentation

This repo is the official pytorch implementation of the CLIPer.

:fire: News

We have released the sourse code of CLIPer (all the models needed will be downloaded automatically when running the code).

Introduction

We introduce the early-layer fusion, including patch embeddings and attention maps of early layer to the final layer of CLIP for improving spatial representation.
We further introduce the fine-grained compensation, utilizing the detailed spacial information in the self-attention maps of Stable Diffusion to obtain the precise segmentation results.
The proposed CLIPer achieve the state-of-the-art performance on multiple datasets in open-vocabulary semantic segmentation tasks. our proposed CLIPer obtains mIoU scores of 69.8% on VOC and 43.3% on Object when using ViT-L/14 backbone.

For further details, please check out our paper.

Installation

Please follow the code bellow to create the environment

conda create -n CLIPer python=3.9
conda activate CLIPer
pip install -r requirement.txt

Data Preparation

Please struct the datasets as follows

datasets
├── ADEchallengeData2016
│   ├── images
│   │   ├── training
│   │   │   ├── ADE_train_00000001.jpg
│   │   ├── validation
│   │   │   ├── ADE_val_00000001.jpg
│   ├── annotations
│   │   ├── training
│   │   │   ├── ADE_train_00000001.png
│   │   ├── validation
│   │   │   ├── ADE_val_00000001.png
├── coco2014
│   ├── train2014
│   │   ├── COCO_train2014_000000000009.jpg
│   ├── val2014
│   │   ├── COCO_val2014_000000000042.jpg
│   ├── coco_seg_anno
│   │   ├── 000000000009.png
├── coco2017
│   ├── train2017
│   │   ├── 000000000009.jpg
│   ├── val2017
│   │   ├── 000000000139.jpg
│   ├── stuff_anno164
│   │   ├── train2017
│   │   │   ├── 000000000009.png
│   │   ├── val2017
│   │   │   ├── 000000000139.png
├── VOCdevkit
│   ├── VOC2010
│   │   ├── JPEGImages
│   │   │   ├── 2007_000027.jpg
│   │   ├── SegmentationClassContext
│   │   │   ├──2008_000002.png
│   ├── VOC2012
│   │   ├── JPEGImages
│   │   │   ├── 2007_000027.jpg
│   │   ├── SegmentationClassAug
│   │   │   ├──2007_000032.png

Evaluation

To evaluate our CLIPer, please enter the scripts folder and run the code

# select the config file to evaluate the code
# evaluate voc dataset with background
sh sh_ovs.sh ../scripts/config/vit-l-14/ovs_voc21.yaml
# evaluate voc dataset without background
sh sh_ovs.sh ../scripts/config/vit-l-14/ovs_voc20.yaml
# ...

Results

Run the code in this repo, you should get similar results (reported in paper are shown in the parentheses) in the following table:

Encoder	VOC	Context	Object	VOC20	Contex59	Stuff	ADE
ViT-B/16	65.9(65.9)	37.6(37.6)	39.3(39.0)	85.4(85.2)	41.7(41.7)	27.5(27.5)	21.4(21.4)
ViT-L/14	70.2(69.8)	38.2(38.0)	43.5(43.3)	90.0(90.0)	43.6(43.6)	29.2(28.7)	24.4(24.4)
ViT-H/14	71.0	39.7	43.2	88.9	44.3	30.7	27.5

Visualization

Citation

@misc{Sun_2024_CLIPer,
      title={CLIPer: Hierarchically Improving Spatial Representation of CLIP for Open-Vocabulary Semantic Segmentation}, 
      author={Lin Sun and Jiale Cao and Jin Xie and Xiaoheng Jiang and Yanwei Pang},
      year={2024},
      eprint={2411.13836},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2411.13836}, 
}

linsun449 / cliper.code

readme