File size: 3,764 Bytes
1915572
 
e2947fb
 
 
 
 
 
 
 
 
 
 
 
1915572
a63b561
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
---
license: mit
tags:
- vision
- image-segmentation
datasets:
- ydshieh/coco_dataset_script
widget:
- src: https://huggingface.co/datasets/shi-labs/oneformer_demo/blob/main/coco.jpeg
  example_title: Person
- src: https://huggingface.co/datasets/shi-labs/oneformer_demo/blob/main/demo_2.jpg
  example_title: Airplane
- src: https://huggingface.co/datasets/shi-labs/oneformer_demo/blob/main/demo.jpeg
  example_title: Corgi
---

# OneFormer

OneFormer model trained on the COCO dataset (large-sized version, Dinat backbone). It was introduced in the paper [OneFormer: One Transformer to Rule Universal Image Segmentation](https://arxiv.org/abs/2211.06220) by Jain et al. and first released in [this repository](https://github.com/SHI-Labs/OneFormer).

![model image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/oneformer_teaser.png)

## Model description

OneFormer is the first multi-task universal image segmentation framework based on transformers. OneFormer needs to be trained only once with a single universal architecture, a single model, and on a single dataset, to outperform existing frameworks across semantic, instance, and panoptic segmentation tasks. OneFormer uses a task token to condition the model on the task in focus, making the architecture task-guided for training, and task-dynamic for inference, all with a single model.

![model image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/oneformer_architecture.png)

## Intended uses & limitations

You can use this particular checkpoint for semantic, instance and panoptic segmentation. See the [model hub](https://huggingface.co/models?search=oneformer) to look for other fine-tuned versions on a different dataset.

### How to use

Here is how to use this model:

```python
from transformers import OneFormerFeatureExtractor, OneFormerForUniversalSegmentation
from PIL import Image
import requests
url = "https://huggingface.co/datasets/shi-labs/oneformer_demo/blob/main/coco.jpeg"
image = Image.open(requests.get(url, stream=True).raw)

# Loading a single model for all three tasks
feature_extractor = OneFormerFeatureExtractor.from_pretrained("shi-labs/oneformer_coco_dinat_large")
model = OneFormerForUniversalSegmentation.from_pretrained("shi-labs/oneformer_coco_dinat_large")

# Semantic Segmentation
semantic_inputs = feature_extractor(images=image, ["semantic"] return_tensors="pt")
semantic_outputs = model(**semantic_inputs)
# pass through feature_extractor for postprocessing
predicted_semantic_map = feature_extractor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]

# Instance Segmentation
instance_inputs = feature_extractor(images=image, ["instance"] return_tensors="pt")
instance_outputs = model(**instance_inputs)
# pass through feature_extractor for postprocessing
predicted_instance_map = feature_extractor.post_process_instance_segmentation(outputs, target_sizes=[image.size[::-1]])[0]["segmentation"]

# Panoptic Segmentation
panoptic_inputs = feature_extractor(images=image, ["panoptic"] return_tensors="pt")
panoptic_outputs = model(**panoptic_inputs)
# pass through feature_extractor for postprocessing
predicted_semantic_map = feature_extractor.post_process_panoptic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]["segmentation"]
```

For more examples, please refer to the [documentation](https://huggingface.co/docs/transformers/master/en/model_doc/oneformer).

### Citation

```bibtex
@article{jain2022oneformer,
      title={{OneFormer: One Transformer to Rule Universal Image Segmentation}},
      author={Jitesh Jain and Jiachen Li and MangTik Chiu and Ali Hassani and Nikita Orlov and Humphrey Shi},
      journal={arXiv}, 
      year={2022}
    }
```