tala-kamel1/fast-depth

markdown Copy

language: en license: apache-2.0 tags: - depth-estimation - fast-depth datasets: - nyu_depth_v2

FastDepth

This repository provides trained models and evaluation code for the FastDepth project at MIT. FastDepth is designed for fast monocular depth estimation on embedded systems.

Model Description

FastDepth is based on a MobileNet-NNConv5 architecture with depthwise separable layers in the decoder, additive skip connections, and network pruning using NetAdapt. It achieves state-of-the-art performance on the NYU Depth V2 dataset while being optimized for real-time inference on embedded devices like the NVIDIA Jetson TX2.

Intended Use

This model is intended for monocular depth estimation from RGB images. It can be used in applications such as:

• Robotics
• Augmented Reality
• Autonomous Driving

How to Use

You can use this model with the Hugging Face transformers library or directly via the Hugging Face API.

Using the API

import requests

API_URL = "https://api-inference.huggingface.co/models/your-username/your-model-name"
headers = {"Authorization": "Bearer YOUR_API_TOKEN"}

def query(filename):
    with open(filename, "rb") as f:
        data = f.read()
    response = requests.post(API_URL, headers=headers, data=data)
    return response.json()

output = query("path_to_image.jpg")
Using Transformers
python
Copy
from transformers import pipeline

depth_estimator = pipeline("depth-estimation", model="your-username/your-model-name")
result = depth_estimator("path_to_image.jpg")
Results
FastDepth achieves the following results on the NYU Depth V2 dataset:

Model	Input Size	MACs [G]	RMSE [m]	delta1	CPU [ms]	GPU [ms]
FastDepth (Pruned)	224×224	0.37	0.604	0.771	37	5.6
<p float="left"> <img src="/tala-kamel1/fast-depth/resolve/main/img/acc_fps_gpu.png" alt="Accuracy vs FPS (GPU)" width="375"> <img src="/tala-kamel1/fast-depth/resolve/main/img/acc_fps_cpu.png" alt="Accuracy vs FPS (CPU)" width="375"> </p>
Citation
If you use this model, please cite the following paper:

bibtex
Copy
@inproceedings{icra_2019_fastdepth,
  author      = {{Wofk, Diana and Ma, Fangchang and Yang, Tien-Ju and Karaman, Sertac and Sze, Vivienne}},
  title       = {{FastDepth: Fast Monocular Depth Estimation on Embedded Systems}},
  booktitle   = {{IEEE International Conference on Robotics and Automation (ICRA)}},
  year        = {{2019}}
}