app.py

#!/usr/bin/env python3
# /// script
# dependencies = [
#     "gradio>=5.0.0",
#     "pillow>=10.0.0",
#     "torch>=2.0.0",
#     "numpy>=1.24.0",
#     "spaces>=0.28.0",
#     "teamwork[diffusers]",
#     "hf-transfer>=0.1.9",
# ]
# [tool.uv.sources]
# teamwork = { git = "https://github.com/samsartor/teamwork" }
# ///
import gradio as gr
import spaces
from teamwork.pipelines import TeamworkPipeline
from PIL import Image
import torch
import numpy as np

pipe = TeamworkPipeline.from_checkpoint(
    "samsartor/teamwork-release",
    "decomposition_heterogeneous_sd3.safetensors",
).to("cuda")

# The outputs you asked for, in display order
OUTPUT_KEYS = [
    "diffuse",
    "specular",
    "roughness",
    "normals",
    "depth",
    "albedo",
    "inverseshading",
    "diffuseshading",
    "residual",
]

def _to_pil(x):
    """Convert pipeline output (PIL / numpy / torch) into a PIL.Image."""
    if x is None:
        return None

    if isinstance(x, Image.Image):
        return x

    if torch.is_tensor(x):
        x = x.detach().float().cpu().numpy()

    if isinstance(x, np.ndarray):
        arr = x
        # handle CHW -> HWC
        if arr.ndim == 3 and arr.shape[0] in (1, 3, 4) and arr.shape[-1] not in (1, 3, 4):
            arr = np.transpose(arr, (1, 2, 0))

        # If float in [0..1] (common), scale to [0..255]
        if arr.dtype != np.uint8:
            arr = np.clip(arr, 0.0, 1.0)
            arr = (arr * 255.0).round().astype(np.uint8)

        # If single channel, make it 'L'
        if arr.ndim == 2:
            return Image.fromarray(arr, mode="L")

        # If HWC with 1 channel, squeeze to L
        if arr.ndim == 3 and arr.shape[2] == 1:
            return Image.fromarray(arr[:, :, 0], mode="L")

        # Otherwise RGB/RGBA
        return Image.fromarray(arr)

    # Fallback: try to coerce
    try:
        return Image.fromarray(x)
    except Exception:
        return None


@spaces.GPU(duration=120)
def process_image(input_image, num_inference_steps=28, guidance_scale=1):
    if input_image is None:
        return [], "Please upload an image."

    if not isinstance(input_image, Image.Image):
        input_image = Image.fromarray(input_image)

    with torch.no_grad():
        result = pipe(
            {"image": input_image},
            num_inference_steps=num_inference_steps,
            guidance_scale=guidance_scale,
        )

    # Build a labeled gallery: [(image, "label"), ...]
    gallery_items = []
    missing = []
    for k in OUTPUT_KEYS:
        if k in result and result[k] is not None:
            img = _to_pil(result[k])
            if img is not None:
                gallery_items.append((img, k))
            else:
                missing.append(k)
        else:
            missing.append(k)

    status = "Successfully processed!"
    if missing:
        status += f" Missing/unreadable keys: {', '.join(missing)}"

    return gallery_items, status


with gr.Blocks(title="Teamwork: Collaborative Diffusion") as demo:
    gr.Markdown(
        """
# 🎨 Teamwork: Collaborative Diffusion

Upload an image to process it using the Teamwork model for intrinsic decomposition.

[Paper](https://bin.samsartor.com/teamwork.pdf) | [Code](https://github.com/samsartor/teamwork) | [Models](https://huggingface.co/samsartor/teamwork-release)
"""
    )

    with gr.Row():
        with gr.Column():
            input_image = gr.Image(label="Input Image", type="pil")
            num_steps = gr.Slider(1, 50, value=28, step=1, label="Inference Steps")
            guidance = gr.Slider(1.0, 10.0, value=1.0, step=0.5, label="Guidance Scale")
            submit_btn = gr.Button("Process Image", variant="primary")

        with gr.Column():
            outputs_gallery = gr.Gallery(
                label="Outputs",
                columns=3,
                height="auto",
                preview=True,
                show_label=True,
            )
            status_text = gr.Textbox(label="Status", interactive=False)

    submit_btn.click(
        fn=process_image,
        inputs=[input_image, num_steps, guidance],
        outputs=[outputs_gallery, status_text],
    )

    gr.Examples(
        examples=[
            ["https://raw.githubusercontent.com/samsartor/teamwork/refs/heads/main/demo/red_glass_sphere.png"],
            ["https://raw.githubusercontent.com/samsartor/teamwork/refs/heads/main/demo/gift_card.png"],
            ["https://raw.githubusercontent.com/samsartor/teamwork/refs/heads/main/demo/pink_input.png"],
        ],
        inputs=input_image,
        label="Example Images",
    )

demo.launch()