depth / app.py
besarismaili's picture
Update app.py
0694a72
import gradio as gr
from transformers import DPTFeatureExtractor, DPTForDepthEstimation
import torch
import numpy as np
from PIL import Image
import os
#import cv2
feature_extractor = DPTFeatureExtractor.from_pretrained("Intel/dpt-large")
model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large")
def get_image_depth(image):
# prepare image for the model
encoding = feature_extractor(image, return_tensors="pt")
# forward pass
with torch.no_grad():
outputs = model(**encoding)
predicted_depth = outputs.predicted_depth
# interpolate to original size
prediction = torch.nn.functional.interpolate(
predicted_depth.unsqueeze(1),
size=image.size[::-1],
mode="bicubic",
align_corners=False,
).squeeze()
output = prediction.cpu().numpy()
formatted = (output * 255 / np.max(output)).astype('uint8')
img = Image.fromarray(formatted)
return img
def process_sequence(files):
file_paths = [file.name for file in files]
for file_path in file_paths:
image = Image.open(file_path)
depth_image = get_image_depth(image)
depth_image.save(os.path.join('output', os.path.basename(file_path)))
return file_paths, gr.Info("This is some info")
title = "# Depth estimation demo"
description = "Demo for Intel's DPT"
with gr.Blocks() as iface:
gr.Markdown(title)
gr.Markdown(description)
with gr.Row():
with gr.Column():
with gr.Tab(label='Singel image'):
image = gr.Image(type="pil")
button = gr.Button(value="Get depth", interactive=True, variant="primary")
image_output=gr.Image(type="pil", label="predicted depth")
with gr.Column():
with gr.Tab(label='Frames'):
file_output = gr.File(visible=False)
upload_button = gr.UploadButton("Select directory", file_types=["image"], file_count="directory")
upload_button.upload(process_sequence, upload_button, file_output)
#output=gr.Video(label="Predicted Depth")
message=gr.Text(value="Check output folder for the depth frames.")
button.click(
fn=get_image_depth,
inputs=[image],
outputs=[image_output]
)
iface.queue(concurrency_count=1)
iface.launch(debug=True, enable_queue=True)