app.py

import cv2
import torch
import numpy as np
from transformers import DPTForDepthEstimation, DPTImageProcessor
import gradio as gr
import torch.quantization
import torch.nn.utils.prune as prune

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = DPTForDepthEstimation.from_pretrained("Intel/dpt-swinv2-tiny-256", torch_dtype=torch.float32)
model.eval()

# Apply global unstructured pruning
parameters_to_prune = [
    (module, "weight") for module in filter(lambda m: isinstance(m, (torch.nn.Conv2d, torch.nn.Linear)), model.modules())
]
prune.global_unstructured(
    parameters_to_prune,
    pruning_method=prune.L1Unstructured,
    amount=0.3,  # Prune 30% of weights
)

for module, _ in parameters_to_prune:
    prune.remove(module, "weight")

# Apply quantization after pruning
model = torch.quantization.quantize_dynamic(
    model, {torch.nn.Linear, torch.nn.Conv2d}, dtype=torch.qint8
).to(device)

processor = DPTImageProcessor.from_pretrained("Intel/dpt-swinv2-tiny-256")

color_map = cv2.applyColorMap(np.arange(256, dtype=np.uint8), cv2.COLORMAP_INFERNO)

input_tensor = torch.zeros((1, 3, 128, 128), dtype=torch.float32, device=device)

def preprocess_image(image):
    return cv2.resize(image, (128, 72), interpolation=cv2.INTER_AREA).transpose(2, 0, 1).astype(np.float32) / 255.0

@torch.inference_mode()
def process_frame(image):
    if image is None:
        return None
    preprocessed = preprocess_image(image)
    input_tensor = torch.from_numpy(preprocessed).unsqueeze(0).to(device)
    
    predicted_depth = model(input_tensor).predicted_depth
    depth_map = predicted_depth.squeeze().cpu().numpy()
    depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
    depth_map = (depth_map * 255).astype(np.uint8)
    depth_map_colored = cv2.applyColorMap(depth_map, cv2.COLORMAP_INFERNO)
    
    return cv2.cvtColor(depth_map_colored, cv2.COLOR_BGR2RGB)

interface = gr.Interface(
    fn=process_frame,
    inputs=gr.Image(sources="webcam", streaming=True),
    outputs="image",
    live=True
)

interface.launch()