app.py

import os

# Disable JIT
os.environ["PYTORCH_JIT"] = "0"

from einops import rearrange
import gradio as gr 
import spaces
import torch  
import torch.nn.functional as F
from PIL import Image, ImageOps
from transformers import AutoModel, CLIPImageProcessor

hf_repo = "nvidia/RADIO-L"

image_processor = CLIPImageProcessor.from_pretrained(hf_repo)
model = AutoModel.from_pretrained(hf_repo, trust_remote_code=True)
model.eval()


title = """RADIO: Reduce All Domains Into One"""
description = """
# RADIO

AM-RADIO is a framework to distill Large Vision Foundation models into a single one.
RADIO, a new vision foundation model, excels across visual domains, serving as a superior replacement for vision backbones.
Integrating CLIP variants, DINOv2, and SAM through distillation, it preserves unique features like text grounding and segmentation correspondence.
Outperforming teachers in ImageNet zero-shot (+6.8%), kNN (+2.39%), and linear probing segmentation (+3.8%) and vision-language models (LLaVa 1.5 up to 1.5%), it scales to any resolution, supports non-square images.

# Instructions

Simply paste an image or pick one from the gallery of examples and then click the "Submit" button.
"""

inputs = [
    gr.Image(type="pil")
]

examples = [
    "samples/IMG_0996.jpeg",
    "samples/IMG_1061.jpeg",
    "samples/IMG_1338.jpeg",
    "samples/IMG_4319.jpeg",
    "samples/IMG_5104.jpeg",
    "samples/IMG_5139.jpeg",
    "samples/IMG_6225.jpeg",
    "samples/IMG_6814.jpeg",
    "samples/IMG_7459.jpeg",
    "samples/IMG_7577.jpeg",
    "samples/IMG_7687.jpeg",
    "samples/IMG_9862.jpeg",
]

outputs = [
    gr.Textbox(label="Feature Shape"),
    gr.Image(),
]

def get_robust_pca(features: torch.Tensor, m: float = 2, remove_first_component=False):
    # features: (N, C)
    # m: a hyperparam controlling how many std dev outside for outliers
    assert len(features.shape) == 2, "features should be (N, C)"
    reduction_mat = torch.pca_lowrank(features, q=3, niter=20)[2]
    colors = features @ reduction_mat
    if remove_first_component:
        colors_min = colors.min(dim=0).values
        colors_max = colors.max(dim=0).values
        tmp_colors = (colors - colors_min) / (colors_max - colors_min)
        fg_mask = tmp_colors[..., 0] < 0.2
        reduction_mat = torch.pca_lowrank(features[fg_mask], q=3, niter=20)[2]
        colors = features @ reduction_mat
    else:
        fg_mask = torch.ones_like(colors[:, 0]).bool()
    d = torch.abs(colors[fg_mask] - torch.median(colors[fg_mask], dim=0).values)
    mdev = torch.median(d, dim=0).values
    s = d / mdev
    try:
        rins = colors[fg_mask][s[:, 0] < m, 0]
        gins = colors[fg_mask][s[:, 1] < m, 1]
        bins = colors[fg_mask][s[:, 2] < m, 2]
        rgb_min = torch.tensor([rins.min(), gins.min(), bins.min()])
        rgb_max = torch.tensor([rins.max(), gins.max(), bins.max()])
    except:
        rins = colors
        gins = colors
        bins = colors
        rgb_min = torch.tensor([rins.min(), gins.min(), bins.min()])
        rgb_max = torch.tensor([rins.max(), gins.max(), bins.max()])

    return reduction_mat, rgb_min.to(reduction_mat), rgb_max.to(reduction_mat)


def get_pca_map(
    feature_map: torch.Tensor,
    img_size,
    interpolation="bicubic",
    return_pca_stats=False,
    pca_stats=None,
):
    """
    feature_map: (1, h, w, C) is the feature map of a single image.
    """
    if feature_map.shape[0] != 1:
        # make it (1, h, w, C)
        feature_map = feature_map[None]
    if pca_stats is None:
        reduct_mat, color_min, color_max = get_robust_pca(
            feature_map.reshape(-1, feature_map.shape[-1])
        )
    else:
        reduct_mat, color_min, color_max = pca_stats
    pca_color = feature_map @ reduct_mat
    pca_color = (pca_color - color_min) / (color_max - color_min)
    pca_color = pca_color.clamp(0, 1)
    pca_color = F.interpolate(
        pca_color.permute(0, 3, 1, 2),
        size=img_size,
        mode=interpolation,
    ).permute(0, 2, 3, 1)
    pca_color = pca_color.cpu().numpy().squeeze(0)
    if return_pca_stats:
        return pca_color, (reduct_mat, color_min, color_max)
    return pca_color


def pad_image_to_multiple_of_16(image):
    # Calculate the new dimensions to make them multiples of 16
    width, height = image.size
    new_width = (width + 15) // 16 * 16
    new_height = (height + 15) // 16 * 16

    # Calculate the padding needed on each side
    pad_width = new_width - width
    pad_height = new_height - height

    left = pad_width // 2
    right = pad_width - left
    top = pad_height // 2
    bottom = pad_height - top

    # Apply the padding
    padded_image = ImageOps.expand(image, (left, top, right, bottom), fill='black')

    return padded_image


@spaces.GPU 
def infer_radio(image):
    """Define the function to generate the output."""
    model.cuda()
    image=pad_image_to_multiple_of_16(image)
    width, height = image.size
    pixel_values = image_processor(images=image, return_tensors='pt').pixel_values
    pixel_values = pixel_values.to(torch.bfloat16).cuda()
    
    _, features = model(pixel_values)
    
    
    num_rows = height // model.patch_size
    num_cols = width // model.patch_size
    
    features = features.detach()
    features = rearrange(features, 'b (h w) c -> b h w c', h=num_rows, w=num_cols).float()
    
    pca_viz = get_pca_map(features, (height, width), interpolation='bilinear')
      
    return f"{features.shape}", pca_viz  


# Create the Gradio interface
demo = gr.Interface(
    fn=infer_radio,
    inputs=inputs,
    examples=examples,
    outputs=outputs,
    title=title,
    description=description
)
  
if __name__ == "__main__":  
    demo.launch()