Create README.md

README.md

@@ -0,0 +1,82 @@
+---
+license: apache-2.0
+---
+
+### Load PixCell-256-Cell-ControlNet model
+
+```python
+import torch
+
+from diffusers import DiffusionPipeline
+from diffusers import AutoencoderKL
+
+device = torch.device('cuda')
+
+# We do not host the weights of the SD3 VAE -- load it from StabilityAI
+sd3_vae = AutoencoderKL.from_pretrained("stabilityai/stable-diffusion-3.5-large", subfolder="vae")
+
+pipeline = DiffusionPipeline.from_pretrained(
+    "StonyBrook-CVLab/PixCell-256-Cell-ControlNet",
+    vae=sd3_vae,
+    custom_pipeline="StonyBrook-CVLab/PixCell-pipeline-ControlNet",
+    trust_remote_code=True,
+)
+
+pipeline.to(device);
+```
+
+### Load [[UNI-2h]](https://huggingface.co/MahmoodLab/UNI2-h) for conditioning
+```python
+import timm
+from timm.data import resolve_data_config
+from timm.data.transforms_factory import create_transform
+
+timm_kwargs = {
+            'img_size': 224, 
+            'patch_size': 14, 
+            'depth': 24,
+            'num_heads': 24,
+            'init_values': 1e-5, 
+            'embed_dim': 1536,
+            'mlp_ratio': 2.66667*2,
+            'num_classes': 0, 
+            'no_embed_class': True,
+            'mlp_layer': timm.layers.SwiGLUPacked, 
+            'act_layer': torch.nn.SiLU, 
+            'reg_tokens': 8, 
+            'dynamic_img_size': True
+        }
+uni_model = timm.create_model("hf-hub:MahmoodLab/UNI2-h", pretrained=True, **timm_kwargs)
+transform = create_transform(**resolve_data_config(uni_model.pretrained_cfg, model=uni_model))
+uni_model.eval()
+uni_model.to(device);
+```
+
+
+### Mask-conditioned generation
+```python
+# Load image
+import numpy as np
+from PIL import Image
+from huggingface_hub import hf_hub_download
+
+# This is an example image/mask pair we provide
+image_path = hf_hub_download(repo_id="StonyBrook-CVLab/PixCell-256-Cell-ControlNet", filename="test_image.png")
+mask_path = hf_hub_download(repo_id="StonyBrook-CVLab/PixCell-256-Cell-ControlNet", filename="test_mask.png")
+image = Image.open(image_path).convert("RGB")
+mask = np.asarray(Image.open(mask_path).convert("RGB"))
+
+# Extract UNI embedding from the image
+uni_inp = uni_transforms(image).unsqueeze(dim=0)
+with torch.inference_mode():
+    uni_emb = uni_model(uni_inp.to(device))
+
+# reshape UNI to (bs, 1, D)
+uni_emb = uni_emb.unsqueeze(1)
+print("Extracted UNI:", uni_emb.shape)
+
+# Get unconditional embedding for classifier-free guidance
+uncond = pipeline_controlnet.get_unconditional_embedding(uni_emb.shape[0])
+# Generate new samples using the given mask
+samples = pipeline_controlnet(uni_embeds=uni_emb, controlnet_input=mask, negative_uni_embeds=uncond, guidance_scale=2.5, num_images_per_prompt=1).images
+```