| 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. |
|
|
