Hugging Face's logo

zeromodels
/
segformer_b5_cityscapes_1024

Image Segmentation
ZeroModels
Keras
PyTorch
JAX
TensorFlow
segformer
semantic-segmentation
cityscapes
Model card Files
xet
 Community
IMvision12 commited on
Commit
dc9516d
·
verified ·
1 Parent(s): 5b47d97

Migrate to zeromodels (rename kf_*.json -> zm_*.json, fix refs in config + README, ensure tag + badge)

Browse files
Files changed (3) hide show
  1. README.md +25 -25
  2. kf_config.json → zm_config.json +28 -28
  3. kf_preprocessor.json → zm_preprocessor.json +27 -27
README.md CHANGED
@@ -3,10 +3,10 @@ pipeline_tag: image-segmentation
3
  license: other
4
  license_link: https://github.com/NVlabs/SegFormer/blob/master/LICENSE
5
  base_model: nvidia/segformer-b5-finetuned-cityscapes-1024-1024
6
- library_name: kerasformers
7
  tags:
8
  - keras
9
- - kerasformers
10
  - segformer
11
  - semantic-segmentation
12
  - image-segmentation
@@ -17,13 +17,13 @@ tags:
17
  - tf
18
  ---
19
 
20
- ## ***See [our collection](https://huggingface.co/collections/kerasformers/segformer-6a6a8b9fc36ea6adef0c1616) for all versions of SegFormer.***
21
 
22
  # Run SegFormer with Keras 3: JAX, PyTorch, or TensorFlow
23
 
24
- [![GitHub](https://img.shields.io/badge/GitHub-KerasFormers-black?logo=github)](https://github.com/IMvision12/KerasFormers) [![Docs](https://img.shields.io/badge/Docs-SegFormer-blue)](https://imvision12.github.io/KerasFormers/segformer/) [![Collection](https://img.shields.io/badge/HF-SegFormer%20collection-yellow)](https://huggingface.co/collections/kerasformers/segformer-6a6a8b9fc36ea6adef0c1616)
25
 
26
- # kerasformers/segformer_b5_cityscapes_1024
27
 
28
  Paper: [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (arXiv:2105.15203)](https://arxiv.org/abs/2105.15203) · [HF Papers](https://huggingface.co/papers/2105.15203)
29
 
@@ -31,7 +31,7 @@ SegFormer pairs a hierarchical transformer encoder (MiT) with a lightweight all-
31
 
32
  For more details on the model, please go to NVIDIA's original [model card](https://huggingface.co/nvidia/segformer-b5-finetuned-cityscapes-1024-1024).
33
 
34
- Pure-**Keras 3** conversion of [`nvidia/segformer-b5-finetuned-cityscapes-1024-1024`](https://huggingface.co/nvidia/segformer-b5-finetuned-cityscapes-1024-1024) for [kerasformers](https://github.com/IMvision12/KerasFormers). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
35
 
36
  This is a **semantic segmentation** checkpoint (`SegFormerSemanticSegment`) for **Cityscapes** (19 classes, MiT-B5, 1024px).
37
 
@@ -42,13 +42,13 @@ import os
42
  os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
43
 
44
  from PIL import Image
45
- from kerasformers.models.segformer import (
46
  SegFormerSemanticSegment,
47
  SegFormerImageProcessor,
48
  )
49
 
50
- model = SegFormerSemanticSegment.from_weights("kerasformers/segformer_b5_cityscapes_1024")
51
- processor = SegFormerImageProcessor.from_weights("kerasformers/segformer_b5_cityscapes_1024")
52
 
53
  image = Image.open("your_image.jpg").convert("RGB")
54
  output = model(processor(image)["pixel_values"], training=False)
@@ -58,29 +58,29 @@ result = processor.post_process_semantic_segmentation(
58
  print(result["unique_classes"], result["class_names"])
59
  ```
60
 
61
- Load any SegFormer variant the same way with `from_weights("kerasformers/<variant>")`:
62
 
63
  | Variant | Hub | Dataset | Res |
64
  |---|---|---|---|
65
- | `segformer_b0_ade_512` | [`kerasformers/segformer_b0_ade_512`](https://huggingface.co/kerasformers/segformer_b0_ade_512) | ADE20K | 512 |
66
- | `segformer_b1_ade_512` | [`kerasformers/segformer_b1_ade_512`](https://huggingface.co/kerasformers/segformer_b1_ade_512) | ADE20K | 512 |
67
- | `segformer_b2_ade_512` | [`kerasformers/segformer_b2_ade_512`](https://huggingface.co/kerasformers/segformer_b2_ade_512) | ADE20K | 512 |
68
- | `segformer_b3_ade_512` | [`kerasformers/segformer_b3_ade_512`](https://huggingface.co/kerasformers/segformer_b3_ade_512) | ADE20K | 512 |
69
- | `segformer_b4_ade_512` | [`kerasformers/segformer_b4_ade_512`](https://huggingface.co/kerasformers/segformer_b4_ade_512) | ADE20K | 512 |
70
- | `segformer_b5_ade_640` | [`kerasformers/segformer_b5_ade_640`](https://huggingface.co/kerasformers/segformer_b5_ade_640) | ADE20K | 640 |
71
- | `segformer_b0_cityscapes_768` | [`kerasformers/segformer_b0_cityscapes_768`](https://huggingface.co/kerasformers/segformer_b0_cityscapes_768) | Cityscapes | 768 |
72
- | `segformer_b0_cityscapes_1024` | [`kerasformers/segformer_b0_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b0_cityscapes_1024) | Cityscapes | 1024 |
73
- | `segformer_b1_cityscapes_1024` | [`kerasformers/segformer_b1_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b1_cityscapes_1024) | Cityscapes | 1024 |
74
- | `segformer_b2_cityscapes_1024` | [`kerasformers/segformer_b2_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b2_cityscapes_1024) | Cityscapes | 1024 |
75
- | `segformer_b3_cityscapes_1024` | [`kerasformers/segformer_b3_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b3_cityscapes_1024) | Cityscapes | 1024 |
76
- | `segformer_b4_cityscapes_1024` | [`kerasformers/segformer_b4_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b4_cityscapes_1024) | Cityscapes | 1024 |
77
- | `segformer_b5_cityscapes_1024` | [`kerasformers/segformer_b5_cityscapes_1024`](https://huggingface.co/kerasformers/segformer_b5_cityscapes_1024) | Cityscapes | 1024 |
78
 
79
  ## Tips
80
 
81
- - Set `KERAS_BACKEND` **before** importing Keras / kerasformers.
82
  - Prefer `SegFormerImageProcessor.from_weights(...)` so the resize matches the variant (ADE B5 is 640; Cityscapes is often 1024).
83
- - See [SegFormer docs](https://imvision12.github.io/KerasFormers/segformer/) and [Loading Weights](https://imvision12.github.io/KerasFormers/loading_weights/).
84
  - Community / upstream safetensors still work via the `hf:` prefix, e.g. `SegFormerSemanticSegment.from_weights("hf:nvidia/segformer-b5-finetuned-cityscapes-1024-1024")`.
85
 
86
  ## Special Thanks
 
3
  license: other
4
  license_link: https://github.com/NVlabs/SegFormer/blob/master/LICENSE
5
  base_model: nvidia/segformer-b5-finetuned-cityscapes-1024-1024
6
+ library_name: zeromodels
7
  tags:
8
  - keras
9
+ - zeromodels
10
  - segformer
11
  - semantic-segmentation
12
  - image-segmentation
 
17
  - tf
18
  ---
19
 
20
+ ## ***See [our collection](https://huggingface.co/collections/zeromodels/segformer-6a6a8b9fc36ea6adef0c1616) for all versions of SegFormer.***
21
 
22
  # Run SegFormer with Keras 3: JAX, PyTorch, or TensorFlow
23
 
24
+ [![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-black?logo=github)](https://github.com/IMvision12/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-SegFormer-blue)](https://imvision12.github.io/ZeroModels/segformer/) [![Collection](https://img.shields.io/badge/HF-SegFormer%20collection-yellow)](https://huggingface.co/collections/zeromodels/segformer-6a6a8b9fc36ea6adef0c1616)
25
 
26
+ # zeromodels/segformer_b5_cityscapes_1024
27
 
28
  Paper: [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (arXiv:2105.15203)](https://arxiv.org/abs/2105.15203) · [HF Papers](https://huggingface.co/papers/2105.15203)
29
 
 
31
 
32
  For more details on the model, please go to NVIDIA's original [model card](https://huggingface.co/nvidia/segformer-b5-finetuned-cityscapes-1024-1024).
33
 
34
+ Pure-**Keras 3** conversion of [`nvidia/segformer-b5-finetuned-cityscapes-1024-1024`](https://huggingface.co/nvidia/segformer-b5-finetuned-cityscapes-1024-1024) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**.
35
 
36
  This is a **semantic segmentation** checkpoint (`SegFormerSemanticSegment`) for **Cityscapes** (19 classes, MiT-B5, 1024px).
37
 
 
42
  os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
43
 
44
  from PIL import Image
45
+ from zeromodels.models.segformer import (
46
  SegFormerSemanticSegment,
47
  SegFormerImageProcessor,
48
  )
49
 
50
+ model = SegFormerSemanticSegment.from_weights("zeromodels/segformer_b5_cityscapes_1024")
51
+ processor = SegFormerImageProcessor.from_weights("zeromodels/segformer_b5_cityscapes_1024")
52
 
53
  image = Image.open("your_image.jpg").convert("RGB")
54
  output = model(processor(image)["pixel_values"], training=False)
 
58
  print(result["unique_classes"], result["class_names"])
59
  ```
60
 
61
+ Load any SegFormer variant the same way with `from_weights("zeromodels/<variant>")`:
62
 
63
  | Variant | Hub | Dataset | Res |
64
  |---|---|---|---|
65
+ | `segformer_b0_ade_512` | [`zeromodels/segformer_b0_ade_512`](https://huggingface.co/zeromodels/segformer_b0_ade_512) | ADE20K | 512 |
66
+ | `segformer_b1_ade_512` | [`zeromodels/segformer_b1_ade_512`](https://huggingface.co/zeromodels/segformer_b1_ade_512) | ADE20K | 512 |
67
+ | `segformer_b2_ade_512` | [`zeromodels/segformer_b2_ade_512`](https://huggingface.co/zeromodels/segformer_b2_ade_512) | ADE20K | 512 |
68
+ | `segformer_b3_ade_512` | [`zeromodels/segformer_b3_ade_512`](https://huggingface.co/zeromodels/segformer_b3_ade_512) | ADE20K | 512 |
69
+ | `segformer_b4_ade_512` | [`zeromodels/segformer_b4_ade_512`](https://huggingface.co/zeromodels/segformer_b4_ade_512) | ADE20K | 512 |
70
+ | `segformer_b5_ade_640` | [`zeromodels/segformer_b5_ade_640`](https://huggingface.co/zeromodels/segformer_b5_ade_640) | ADE20K | 640 |
71
+ | `segformer_b0_cityscapes_768` | [`zeromodels/segformer_b0_cityscapes_768`](https://huggingface.co/zeromodels/segformer_b0_cityscapes_768) | Cityscapes | 768 |
72
+ | `segformer_b0_cityscapes_1024` | [`zeromodels/segformer_b0_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b0_cityscapes_1024) | Cityscapes | 1024 |
73
+ | `segformer_b1_cityscapes_1024` | [`zeromodels/segformer_b1_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b1_cityscapes_1024) | Cityscapes | 1024 |
74
+ | `segformer_b2_cityscapes_1024` | [`zeromodels/segformer_b2_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b2_cityscapes_1024) | Cityscapes | 1024 |
75
+ | `segformer_b3_cityscapes_1024` | [`zeromodels/segformer_b3_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b3_cityscapes_1024) | Cityscapes | 1024 |
76
+ | `segformer_b4_cityscapes_1024` | [`zeromodels/segformer_b4_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b4_cityscapes_1024) | Cityscapes | 1024 |
77
+ | `segformer_b5_cityscapes_1024` | [`zeromodels/segformer_b5_cityscapes_1024`](https://huggingface.co/zeromodels/segformer_b5_cityscapes_1024) | Cityscapes | 1024 |
78
 
79
  ## Tips
80
 
81
+ - Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
82
  - Prefer `SegFormerImageProcessor.from_weights(...)` so the resize matches the variant (ADE B5 is 640; Cityscapes is often 1024).
83
+ - See [SegFormer docs](https://imvision12.github.io/ZeroModels/segformer/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/).
84
  - Community / upstream safetensors still work via the `hf:` prefix, e.g. `SegFormerSemanticSegment.from_weights("hf:nvidia/segformer-b5-finetuned-cityscapes-1024-1024")`.
85
 
86
  ## Special Thanks
kf_config.json → zm_config.json RENAMED
@@ -1,29 +1,29 @@
1
- {
2
- "library_name": "kerasformers",
3
- "kerasformers_version": "1.2.1",
4
- "model_module": "kerasformers.models.segformer",
5
- "model_class": "SegFormerSemanticSegment",
6
- "variant": "segformer_b5_cityscapes_1024",
7
- "weights": "model.weights.h5",
8
- "schema_version": 2,
9
- "weight_dtype": "float32",
10
- "model_type": "segformer",
11
- "vision_config": {
12
- "embed_dim": [
13
- 64,
14
- 128,
15
- 320,
16
- 512
17
- ],
18
- "depths": [
19
- 3,
20
- 6,
21
- 40,
22
- 3
23
- ],
24
- "decode_head_dim": 768,
25
- "dropout_rate": 0.1,
26
- "num_classes": 19,
27
- "image_size": 1024
28
- }
29
  }
 
1
+ {
2
+ "library_name": "zeromodels",
3
+ "zeromodels_version": "1.2.1",
4
+ "model_module": "zeromodels.models.segformer",
5
+ "model_class": "SegFormerSemanticSegment",
6
+ "variant": "segformer_b5_cityscapes_1024",
7
+ "weights": "model.weights.h5",
8
+ "schema_version": 2,
9
+ "weight_dtype": "float32",
10
+ "model_type": "segformer",
11
+ "vision_config": {
12
+ "embed_dim": [
13
+ 64,
14
+ 128,
15
+ 320,
16
+ 512
17
+ ],
18
+ "depths": [
19
+ 3,
20
+ 6,
21
+ 40,
22
+ 3
23
+ ],
24
+ "decode_head_dim": 768,
25
+ "dropout_rate": 0.1,
26
+ "num_classes": 19,
27
+ "image_size": 1024
28
+ }
29
  }
kf_preprocessor.json → zm_preprocessor.json RENAMED
@@ -1,28 +1,28 @@
1
- {
2
- "library_name": "kerasformers",
3
- "kerasformers_version": "1.1.3",
4
- "preprocessor_module": "kerasformers.models.segformer",
5
- "preprocessor_class": "SegFormerImageProcessor",
6
- "variant": null,
7
- "do_resize": true,
8
- "size": {
9
- "height": 1024,
10
- "width": 1024
11
- },
12
- "resample": "bilinear",
13
- "do_rescale": true,
14
- "rescale_factor": 0.00392156862745098,
15
- "do_normalize": true,
16
- "image_mean": [
17
- 0.485,
18
- 0.456,
19
- 0.406
20
- ],
21
- "image_std": [
22
- 0.229,
23
- 0.224,
24
- 0.225
25
- ],
26
- "return_tensor": true,
27
- "data_format": null
28
  }
 
1
+ {
2
+ "library_name": "zeromodels",
3
+ "zeromodels_version": "1.1.3",
4
+ "preprocessor_module": "zeromodels.models.segformer",
5
+ "preprocessor_class": "SegFormerImageProcessor",
6
+ "variant": null,
7
+ "do_resize": true,
8
+ "size": {
9
+ "height": 1024,
10
+ "width": 1024
11
+ },
12
+ "resample": "bilinear",
13
+ "do_rescale": true,
14
+ "rescale_factor": 0.00392156862745098,
15
+ "do_normalize": true,
16
+ "image_mean": [
17
+ 0.485,
18
+ 0.456,
19
+ 0.406
20
+ ],
21
+ "image_std": [
22
+ 0.229,
23
+ 0.224,
24
+ 0.225
25
+ ],
26
+ "return_tensor": true,
27
+ "data_format": null
28
  }