DepthART Relative-Depth Model
=============================

This repository contains the DepthART relative-depth model, converted into
the `.depthart` binary container format used by the TypeGPU runtime.

Original work
-------------
Title:      DepthART: Scaling Foundation Monocular Depth to Tiny Models
Authors:    Feng Xue, Wu Chen, Mingshuai Zhao, Guofeng Zhong, Anlong Ming,
            Haozhe Wang, Dianqiao Lei, Zhaowen Lin, Haiyang Zhang, Nicu Sebe
Published:  ACM Multimedia, 2026
Project:    https://github.com/xuefeng-cvr/DepthART
Model host: https://huggingface.co/Fengxue93/DepthART
License:    Apache License, Version 2.0

The upstream project repository does not carry a LICENSE file or a copyright
notice. Apache License, Version 2.0 is the license declared by the model
repository metadata at the model host above, and is the license under which
this asset is redistributed.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

Modifications
-------------
The original checkpoint has been converted into the `.depthart` binary
container format compatible with the TypeGPU runtime. Batch normalization and
reparameterizable branches were folded, graph-only view operations were
lowered, and compatible activations and channel-affine operations were fused.

The computational behavior has been altered: selected encoder and decoder
weights and activations are stored and computed in FP16. Selective-scan
recurrence, normalization, residual endpoints, public input and output, and
quality-sensitive reconstruction paths remain FP32. Model output therefore
differs numerically from the original checkpoint.

No upstream source code is included in this asset. The TypeGPU inference
implementation is an independent implementation written for this project.

This conversion was performed by Software Mansion S.A. as part of the TypeGPU
project (https://github.com/software-mansion/TypeGPU). The converter and the
exact reproduction command are in `tools/depthart` in that repository.

Affiliation
-----------
This is an unofficial conversion. It is not affiliated with, endorsed by, or
sponsored by the DepthART authors or their institutions. References to
DepthART herein are made for the sole purpose of identifying the upstream
source of the model in accordance with Apache License 2.0 attribution
requirements.