DepthHist: Depth Estimation Using Differentiable Histogram with SliceStitch Augmentation

---

DepthHist is a depth estimation model with a trainable histogram-like layer that dynamically represents depth distributions. It optimizes bin centers and dispersion factors to encode depth as a continuous distribution, with two differentiable histogram-based loss functions to enforce global alignment and structural coherence.

📝 Table of Contents

• About
• Getting Started
• Installation
• Data Download
• Training and Evaluation
• Model Checkpoints
• Author

🧐 About

DepthHist estimates depth by combining a Swin Transformer-based feature extractor with a U-Net-inspired decoder. The key component, HistLayer, integrates RGB input, encoder features, and upsampled decoder outputs to dynamically represent depth distributions. The final depth map is obtained through element-wise multiplication and cumulative summation of HistLayer outputs.

🏁 Getting Started

These instructions will help you set up and run the project on your local machine for training and evaluation purposes.

Prerequisites

Ensure you have Git installed, then clone the repository:

git clone https://github.com/Chaouki-AI/DepthHist
cd DepthHist/

📦 Installation

Ensure you have Anaconda installed on your machine. Then, run the following command to set up the environment:

chmod +x ./setup/installEnv.sh
./setup/installEnv.sh
conda activate DepthHist

📂 Data Download

Follow the instructions in the LIDAR-DATABOX repository to download datasets such as KITTI, NYUv2, and others used for training and evaluation of depth estimation models.

🚀 Training and Evaluation

Training

To train the model, update the args.txt file (examples can be found in the args/train folder). Then, start training with:

python main.py @args.txt

In case of training multiple model variations, you can use run.sh.

First, define the various configurations using different args_conf1.txt files, add them to the script, and then run:

chmod +x run.sh
./run.sh 

By default, Checkpoints will be saved in the checkpoints folder, and TensorBoard logs will be stored in the runs folder. You can change it using the args.txt file.

Evaluation

To evaluate the model, create an args.txt file as provided in the args/eval/ folder (e.g., eval_args_nyu_bins_cauchy.txt). Then, run:

python evaluation.py @args/eval/eval_args_nyu_bins_cauchy.txt

The evaluation results will be saved as a .txt file in the Eval_result folder, timestamped for easy identification.

📌 Model Checkpoints

Below are the model checkpoints for the KITTI and NYUv2 datasets, along with their evaluation metrics:

Dataset	a1	a2	a3	Abs Rel	RMSE	Log_10	RMSE Log	SILog	Sq Rel	Checkpoint
KITTI	0.983	0.998	0.999	0.048	1.987	0.021	0.070	6.442	0.119	Download
NYUv2	0.962	0.996	0.999	0.076	0.292	0.032	0.129	8.244	0.03	Download

✍️ Author

M. Chaouki ZIARA is affiliated with the RCAM Laboratory, Department of Electronics, Djillali Liabes University, Sidi Bel Abbes, Algeria. (Email: chaouki.ziara@univ-sba.dz, medchaoukiziara@gmail.com) – concept creator, algorithm development, implementation, and manuscript writing.