EDADepth: Enhanced Data Augmentation for Monocular Depth Estimation

Installation:

Clone the repo and run the following commands:

cd EDADepth_ICMLA
conda env create -f edadepth_env.yml
conda activate EDADepth

Preparing datasets:

Prepare Official NYUDepthv2 dataset following the instructions from BTS. After extracting the datasets, copy the datasets to the following directories:

NYUv2:

EDADepth_ICMLA/depth/data/nyu

KITTI:

EDADepth_ICMLA/depth/data/kitti

The dataset structure should look similar to the following:

nyu
├── nyu_depth_v2
│   ├── official_splits
│   └── sync
├── sync/

kitti
├──data_depth_annotated/
├──raw_data/
├──val_selection_cropped/

Stable Diffusion Checkpoint

Please download the Stable-diffusion-v1-5-eamonly-pruned checkpoint from this link and paste it to the following directory:

EDADepth_ICMLA/checkpoints

Evaluation using Pre-trained models:

If you want to evaluate the test dataset using our pre-trained models, you can download our pre-trained checkpoints.

• Download our pre-trained checkpoint for NYUv2 dataset using this link
• Download our pre-trained checkpoint for KITTI dataset using this link

After downloading the checkpoints, copy them and paste them to the following directory:

EDADepth_ICMLA/depth/checkpoints_depth

Finally, navigate to depth directory and run the following commands:

For NYUv2:

bash test_nyu.sh

For KITTI:

bash test_kitti.sh

Acknolwledgements

Our source code is inspired from ECoDepth, VPD, Stable Diffusion, LDM and GLPDepth