Feature Extraction
Transformers
Safetensors
English
remote-sensing
earth-observation
vision
croma
sentinel-1
sentinel-2
multimodal
Instructions to use BiliSakura/CROMA-transformers with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use BiliSakura/CROMA-transformers with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="BiliSakura/CROMA-transformers")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("BiliSakura/CROMA-transformers", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Sync updated code and configs (no weight re-upload)
Browse files
README.md
CHANGED
|
@@ -1,20 +1,3 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
language:
|
| 4 |
-
- en
|
| 5 |
-
tags:
|
| 6 |
-
- remote-sensing
|
| 7 |
-
- earth-observation
|
| 8 |
-
- vision
|
| 9 |
-
- feature-extraction
|
| 10 |
-
- croma
|
| 11 |
-
- sentinel-1
|
| 12 |
-
- sentinel-2
|
| 13 |
-
- multimodal
|
| 14 |
-
library_name: transformers
|
| 15 |
-
pipeline_tag: feature-extraction
|
| 16 |
-
---
|
| 17 |
-
|
| 18 |
# CROMA Transformers Models
|
| 19 |
|
| 20 |
Self-contained HuggingFace model checkpoints for [CROMA](https://arxiv.org/pdf/2311.00566.pdf). Each subfolder ships its own model, image processor, and custom pipeline code — no external `croma` package required.
|
|
@@ -30,6 +13,8 @@ CROMA expects dual-modality Sentinel-1 (2-channel SAR) and Sentinel-2 (12-channe
|
|
| 30 |
|
| 31 |
## Usage
|
| 32 |
|
|
|
|
|
|
|
| 33 |
```python
|
| 34 |
from transformers import pipeline
|
| 35 |
|
|
@@ -43,8 +28,8 @@ pipe = pipeline(
|
|
| 43 |
|
| 44 |
import numpy as np
|
| 45 |
|
| 46 |
-
sar = np.random.randn(2,
|
| 47 |
-
optical = np.random.randn(12,
|
| 48 |
|
| 49 |
joint_gap = pipe(
|
| 50 |
sar_images=sar,
|
|
@@ -62,6 +47,19 @@ sar_tokens = pipe(
|
|
| 62 |
)
|
| 63 |
```
|
| 64 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
Or load components directly:
|
| 66 |
|
| 67 |
```python
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# CROMA Transformers Models
|
| 2 |
|
| 3 |
Self-contained HuggingFace model checkpoints for [CROMA](https://arxiv.org/pdf/2311.00566.pdf). Each subfolder ships its own model, image processor, and custom pipeline code — no external `croma` package required.
|
|
|
|
| 13 |
|
| 14 |
## Usage
|
| 15 |
|
| 16 |
+
Processors default to **`do_resize: false`**. CROMA expects SAR `(2, H, W)` and optical `(12, H, W)` at the same native spatial size (120×120 at pretraining; larger sizes work without forced resize).
|
| 17 |
+
|
| 18 |
```python
|
| 19 |
from transformers import pipeline
|
| 20 |
|
|
|
|
| 28 |
|
| 29 |
import numpy as np
|
| 30 |
|
| 31 |
+
sar = np.random.randn(2, 256, 256).astype(np.float32)
|
| 32 |
+
optical = np.random.randn(12, 256, 256).astype(np.float32)
|
| 33 |
|
| 34 |
joint_gap = pipe(
|
| 35 |
sar_images=sar,
|
|
|
|
| 47 |
)
|
| 48 |
```
|
| 49 |
|
| 50 |
+
Opt in to 120×120 resize:
|
| 51 |
+
|
| 52 |
+
```python
|
| 53 |
+
joint_gap = pipe(
|
| 54 |
+
sar_images=sar,
|
| 55 |
+
optical_images=optical,
|
| 56 |
+
use_8_bit=True,
|
| 57 |
+
pool=True,
|
| 58 |
+
return_tensors=True,
|
| 59 |
+
image_processor_kwargs={"do_resize": True},
|
| 60 |
+
)
|
| 61 |
+
```
|
| 62 |
+
|
| 63 |
Or load components directly:
|
| 64 |
|
| 65 |
```python
|
croma-base-patch8-120/image_processing_croma.py
CHANGED
|
@@ -113,7 +113,7 @@ class CromaImageProcessor(BaseImageProcessor):
|
|
| 113 |
|
| 114 |
def __init__(
|
| 115 |
self,
|
| 116 |
-
do_resize: bool =
|
| 117 |
size: Optional[dict[str, int]] = None,
|
| 118 |
resample: PILImageResampling = PILImageResampling.BILINEAR,
|
| 119 |
do_rescale: bool = True,
|
|
|
|
| 113 |
|
| 114 |
def __init__(
|
| 115 |
self,
|
| 116 |
+
do_resize: bool = False,
|
| 117 |
size: Optional[dict[str, int]] = None,
|
| 118 |
resample: PILImageResampling = PILImageResampling.BILINEAR,
|
| 119 |
do_rescale: bool = True,
|
croma-base-patch8-120/preprocessor_config.json
CHANGED
|
@@ -4,7 +4,7 @@
|
|
| 4 |
"height": 120,
|
| 5 |
"width": 120
|
| 6 |
},
|
| 7 |
-
"do_resize":
|
| 8 |
"do_rescale": true,
|
| 9 |
"do_normalize": true,
|
| 10 |
"use_8_bit": true,
|
|
@@ -13,4 +13,4 @@
|
|
| 13 |
"auto_map": {
|
| 14 |
"AutoImageProcessor": "image_processing_croma.CromaImageProcessor"
|
| 15 |
}
|
| 16 |
-
}
|
|
|
|
| 4 |
"height": 120,
|
| 5 |
"width": 120
|
| 6 |
},
|
| 7 |
+
"do_resize": false,
|
| 8 |
"do_rescale": true,
|
| 9 |
"do_normalize": true,
|
| 10 |
"use_8_bit": true,
|
|
|
|
| 13 |
"auto_map": {
|
| 14 |
"AutoImageProcessor": "image_processing_croma.CromaImageProcessor"
|
| 15 |
}
|
| 16 |
+
}
|
croma-large-patch8-120/image_processing_croma.py
CHANGED
|
@@ -113,7 +113,7 @@ class CromaImageProcessor(BaseImageProcessor):
|
|
| 113 |
|
| 114 |
def __init__(
|
| 115 |
self,
|
| 116 |
-
do_resize: bool =
|
| 117 |
size: Optional[dict[str, int]] = None,
|
| 118 |
resample: PILImageResampling = PILImageResampling.BILINEAR,
|
| 119 |
do_rescale: bool = True,
|
|
|
|
| 113 |
|
| 114 |
def __init__(
|
| 115 |
self,
|
| 116 |
+
do_resize: bool = False,
|
| 117 |
size: Optional[dict[str, int]] = None,
|
| 118 |
resample: PILImageResampling = PILImageResampling.BILINEAR,
|
| 119 |
do_rescale: bool = True,
|
croma-large-patch8-120/preprocessor_config.json
CHANGED
|
@@ -4,7 +4,7 @@
|
|
| 4 |
"height": 120,
|
| 5 |
"width": 120
|
| 6 |
},
|
| 7 |
-
"do_resize":
|
| 8 |
"do_rescale": true,
|
| 9 |
"do_normalize": true,
|
| 10 |
"use_8_bit": true,
|
|
@@ -13,4 +13,4 @@
|
|
| 13 |
"auto_map": {
|
| 14 |
"AutoImageProcessor": "image_processing_croma.CromaImageProcessor"
|
| 15 |
}
|
| 16 |
-
}
|
|
|
|
| 4 |
"height": 120,
|
| 5 |
"width": 120
|
| 6 |
},
|
| 7 |
+
"do_resize": false,
|
| 8 |
"do_rescale": true,
|
| 9 |
"do_normalize": true,
|
| 10 |
"use_8_bit": true,
|
|
|
|
| 13 |
"auto_map": {
|
| 14 |
"AutoImageProcessor": "image_processing_croma.CromaImageProcessor"
|
| 15 |
}
|
| 16 |
+
}
|