Add usage instructions

README.md

@@ -26,6 +26,45 @@ There are no dedicated datasets for image colorisation, hence I curated my own d
 <img src="https://cdn-uploads.huggingface.co/production/uploads/6318256d212fce5a3cde0fe3/5eKOaiTUK4uDeq07MdJIY.png" width="650px"/>
 <img src="https://cdn-uploads.huggingface.co/production/uploads/6318256d212fce5a3cde0fe3/MI08kcYgav2ouXlHvvNtu.png" width="650px"/>
 
+# Usage
+Download the architecture file and model weights
+```python
+hf_hub_download(
+    repo_id="ayushshah/imagecolorization",
+    filename="model.py",
+    local_dir=".",
+    local_dir_use_symlinks=False
+)
+
+weights_path = hf_hub_download(
+    repo_id=REPO_ID,
+    filename="model.safetensors"
+)
+```
+
+Make sure the input image(s) are of the size 224x224. Convert them to LAB color space. You can use [`kornia`](https://kornia.readthedocs.io/en/stable/color.html#kornia.color.rgb_to_lab).
+Isolate the L channel and make sure it is in the range [0, 1]. L channel is originally in the range [0, 100].
+
+```python
+from model import UNet
+from safetensors.torch import load_file
+
+model = UNet().to(DEVICE)
+state_dict = load_file(weights_path)
+model.load_state_dict(state_dict)
+model.eval()
+
+with torch.no_grad():
+    ab_pred = model(L_normalized)
+```
+
+The outputs are in the range [-1, 1]. You can convert the ab channels to their original range using a linear scaling function. Afterwards you can concatenate the original L and the ab channels to get the LAB image.
+```python
+ab = (ab+1) * 255.0 / 2 - 128.0
+ab = torch.clamp(ab, -128, 127)
+lab = torch.cat((L, ab), dim=1)
+```
+
 # References
 - [Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image Colorization with Simultaneous Classification](https://iizuka.cs.tsukuba.ac.jp/projects/colorization/data/colorization_sig2016.pdf)
 - [Colorful Image Colorization](https://arxiv.org/pdf/1603.08511)