README.md

---
license: mit
model-index:
  - name: CLIP-EBC (ViT-B/16)
    results:
      - task:
          type: crowd-counting
        dataset:
          name: nwpu_crowd
          type: nwpu_crowd
        metrics:
          - name: MAE
            type: MAE
            value: 61.3
          - name: RMSE
            type: RMSE
            value: 278.4
        source:
          name: NWPU-Crowd Benchmark
          url: https://www.crowdbenchmark.com/nwpucrowd.html
  - name: CLIP-EBC (ViT-L/14)
    results:
      - task:
          type: crowd-counting
        dataset:
          name: nwpu_crowd
          type: nwpu_crowd
        metrics:
          - name: MAE
            type: MAE
            value: 58.2
          - name: RMSE
            type: RMSE
            value: 268.5
        source:
          name: NWPU-Crowd Benchmark
          url: https://www.crowdbenchmark.com/nwpucrowd.html
---
# 🚀 CLIP-EBC


[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/clip-ebc-clip-can-count-accurately-through/crowd-counting-on-ucf-qnrf)](https://paperswithcode.com/sota/crowd-counting-on-ucf-qnrf?p=clip-ebc-clip-can-count-accurately-through)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/clip-ebc-clip-can-count-accurately-through/crowd-counting-on-shanghaitech-a)](https://paperswithcode.com/sota/crowd-counting-on-shanghaitech-a?p=clip-ebc-clip-can-count-accurately-through)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/clip-ebc-clip-can-count-accurately-through/crowd-counting-on-shanghaitech-b)](https://paperswithcode.com/sota/crowd-counting-on-shanghaitech-b?p=clip-ebc-clip-can-count-accurately-through)
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/clip-ebc-clip-can-count-accurately-through/crowd-counting-on-nwpu-crowd-val)](https://paperswithcode.com/sota/crowd-counting-on-nwpu-crowd-val?p=clip-ebc-clip-can-count-accurately-through)

The official implementation of **CLIP-EBC**, proposed in the paper [*CLIP-EBC: CLIP Can Count Accurately through Enhanced Blockwise Classification*](https://arxiv.org/abs/2403.09281). 

For more details, please refer to the [GitHub repo](https://github.com/Yiming-M/CLIP-EBC).