ZoeDAM

Depth Anything models focused on robust relative depth estimation. To achieve metric depth estimation, we follow ZoeDepth to fine-tune from Depth Anything pre-trained encoder with metric depth information from NYUv2.

Method	$\delta_1 \uparrow$	$\delta_2 \uparrow$	$\delta_3 \uparrow$	AbsRel $\downarrow$	RMSE $\downarrow$	log10 $\downarrow$
ZoeDepth	0.955	0.995	0.999	0.075	0.270	0.032
ZoeDAM-L	0.984	0.998	1.000	0.056	0.206	0.024
ZoeDAM-B (Ours)	0.976	0.997	1.000	0.061	0.254	0.029

Installation

conda env create -n zoedam python=3.9.7
conda activate zoedam
pip install -r requirements.txt

Please follow ZoeDepth to prepare the training and test datasets.

Training

Please first download our Depth Anything pre-trained model here, and put it under the checkpoints directory.

python train_mono.py -m zoedepth -d nyu --pretrained_resource=""

This will automatically use our Depth Anything pre-trained ViT-L encoder.

Evaluation

Make sure you have downloaded our pre-trained metric-depth models here (for evaluation) and pre-trained relative-depth model here (for initializing the encoder) and put them under the checkpoints directory.

Indoor:

python evaluate.py -m zoedepth --pretrained_resource="local::./checkpoints/depth_anything_metric_depth_indoor.pt" -d <nyu | sunrgbd | ibims | hypersim_test>