Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,64 @@
|
|
| 1 |
-
---
|
| 2 |
-
license:
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
library_name: pytorch
|
| 4 |
+
tags:
|
| 5 |
+
- diffusion
|
| 6 |
+
- counting
|
| 7 |
+
- hallucination
|
| 8 |
+
- image-regression
|
| 9 |
+
- resnet
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# CountHallu — SimObject Counting Model
|
| 13 |
+
|
| 14 |
+
Counting model from **[Counting Hallucinations in Diffusion Models](https://arxiv.org/abs/2510.13080)**
|
| 15 |
+
(arXiv:2510.13080). It scores images generated by a diffusion model trained on
|
| 16 |
+
[SimObject](https://huggingface.co/datasets/ShyFoo/CountHallu-dataset-SimObject) to
|
| 17 |
+
decide whether each sample is counting-correct or a counting hallucination.
|
| 18 |
+
|
| 19 |
+
## Architecture & checkpoint
|
| 20 |
+
|
| 21 |
+
- **ResNet-50** (torchvision, ImageNet-1k V2 init) with the final layer replaced by
|
| 22 |
+
a 3-output regression head — one predicted instance count per object class.
|
| 23 |
+
- Ships a single `model.pth` (a plain `state_dict`).
|
| 24 |
+
- **Decision rule:** round the 3 predictions; a sample is a *hallucination* if any
|
| 25 |
+
class ≥ 2 or all classes are 0 (valid SimObject images have at most one instance
|
| 26 |
+
per class and at least one object).
|
| 27 |
+
|
| 28 |
+
## Usage
|
| 29 |
+
|
| 30 |
+
Inputs are RGB images normalised to `[-1, 1]` (`ToTensor` + `Normalize([0.5]*3,
|
| 31 |
+
[0.5]*3)`).
|
| 32 |
+
|
| 33 |
+
```python
|
| 34 |
+
import torch
|
| 35 |
+
from huggingface_hub import hf_hub_download
|
| 36 |
+
from counthallu.models.counting import CountingRegressor
|
| 37 |
+
|
| 38 |
+
ckpt = hf_hub_download("ShyFoo/CountHallu-counting_model-SimObject", "model.pth")
|
| 39 |
+
model = CountingRegressor(num_classes=3)
|
| 40 |
+
model.load_state_dict(torch.load(ckpt, map_location="cpu"))
|
| 41 |
+
model.eval()
|
| 42 |
+
```
|
| 43 |
+
|
| 44 |
+
Or let the evaluation protocol fetch it for you:
|
| 45 |
+
|
| 46 |
+
```python
|
| 47 |
+
from counthallu.utils import load_counting_model
|
| 48 |
+
model, model_type, _, _ = load_counting_model(
|
| 49 |
+
"simobject", use_hub_model=True,
|
| 50 |
+
repo_id="ShyFoo/CountHallu-counting_model-SimObject")
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
See the [CountHallu repository](<GITHUB_URL>) for the full evaluation protocol.
|
| 54 |
+
|
| 55 |
+
## Citation
|
| 56 |
+
|
| 57 |
+
```bibtex
|
| 58 |
+
@article{fu2025counting,
|
| 59 |
+
title={Counting Hallucinations in Diffusion Models},
|
| 60 |
+
author={Fu, Shuai and Zhou, Jian and Chen, Qi and Jing, Huang and Nguyen, Huy Anh and Liu, Xiaohan and Zeng, Zhixiong and Ma, Lin and Zhang, Quanshi and Wu, Qi},
|
| 61 |
+
journal={arXiv preprint arXiv:2510.13080},
|
| 62 |
+
year={2025}
|
| 63 |
+
}
|
| 64 |
+
```
|