From the Frontier Research Team at Takara.ai we present DepthPro-Safetensors, a memory-efficient and optimized implementation of Apple's high-precision depth estimation model.

---

DepthPro-Safetensors

This repository contains Apple's DepthPro depth estimation model converted to the SafeTensors format for improved memory efficiency, security, and faster loading times.

Model Overview

DepthPro is a state-of-the-art monocular depth estimation model developed by Apple that produces sharp and accurate metric depth maps from a single image in less than a second. This converted version preserves all the capabilities of the original model while providing the benefits of the SafeTensors format.

Technical Specifications

• Total Parameters: 951,991,330
• Memory Usage: 1815.78 MB
• Precision: torch.float16
• Estimated FLOPs: 3,501,896,768

Details calculated with TensorKIKO

Usage

from transformers import AutoModelForDepthEstimation, AutoImageProcessor
import torch
from PIL import Image

# Load model and processor
model = AutoModelForDepthEstimation.from_pretrained("takara-ai/DepthPro-Safetensors")
processor = AutoImageProcessor.from_pretrained("takara-ai/DepthPro-Safetensors")

# Prepare image
image = Image.open("your_image.jpg")
inputs = processor(images=image, return_tensors="pt")

# Inference
with torch.no_grad():
    outputs = model(**inputs)
    predicted_depth = outputs.predicted_depth

# Post-process for visualization
depth_map = processor.post_process_depth_estimation(outputs, target_size=image.size[::-1])

Benefits of SafeTensors Format

• Improved Security: Resistant to code execution vulnerabilities
• Faster Loading Times: Optimized memory mapping for quicker model initialization
• Memory Efficiency: Better handling of tensor storage for reduced memory footprint
• Parallel Loading: Support for efficient parallel tensor loading

Citation

@article{Bochkovskii2024:arxiv,
  author     = {Aleksei Bochkovskii and Ama\"{e}l Delaunoy and Hugo Germain and Marcel Santos and
               Yichao Zhou and Stephan R. Richter and Vladlen Koltun},
  title      = {Depth Pro: Sharp Monocular Metric Depth in Less Than a Second},
  journal    = {arXiv},
  year       = {2024},
}

---

For research inquiries and press, please reach out to [email protected]

人類を変革する