From 8689b487d001d1f49b8580d2404505b9c342b84f Mon Sep 17 00:00:00 2001 From: Jon Froehlich Date: Mon, 3 Aug 2026 15:52:04 -0700 Subject: [PATCH 1/2] README: point paper readers at the v1.0-iccv2025 release main keeps evolving past the paper; the tag (and its GitHub Release, "paper state") is where the as-published version lives, and the HF model and dataset are versioned independently. One callout at the top so a reader landing on an evolving main finds the frozen artifact in one click. Co-Authored-By: Claude Fable 5 --- README.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/README.md b/README.md index 39bbf15..c2f1cf6 100644 --- a/README.md +++ b/README.md @@ -22,6 +22,16 @@

RampNet is a two-stage pipeline that addresses the scarcity of curb ramp detection datasets by using government location data to automatically generate over 210,000 annotated Google Street View panoramas. This new dataset is then used to train a state-of-the-art curb ramp detection model that significantly outperforms previous efforts. In this repo, we provide code for training and testing our system.

+ +> [!NOTE] +> **Using RampNet exactly as the ICCV'25 paper describes it?** Use the tagged release +> [`v1.0-iccv2025`](https://github.com/ProjectSidewalk/RampNet/releases/tag/v1.0-iccv2025) — the +> repository frozen at paper state. The `main` branch keeps evolving (post-paper analyses, +> benchmarks, and groundwork for what comes next), while the published +> [model](https://huggingface.co/projectsidewalk/rampnet-model) and +> [dataset](https://huggingface.co/datasets/projectsidewalk/rampnet-dataset) on Hugging Face are +> versioned independently and match the paper. +
## Citation From 4190e4c498c7aa82731f076bbf9e2efc607baeff Mon Sep 17 00:00:00 2001 From: Jon Froehlich Date: Tue, 4 Aug 2026 05:43:56 -0700 Subject: [PATCH 2/2] README: point at the corrected-eval release too, not just paper state MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The note steered a paper reader at v1.0-iccv2025 — which is exactly the tag whose evaluators the Erratum fifteen lines below says bias precision and recall upward. Someone who reads the note and clicks through never sees that. Two pointers instead of one, and the Erratum linked from the note: paper state for reproducing the paper as written, v1.1-corrected-eval for the same code and weights scored under standard one-to-one matching. Co-Authored-By: Claude Opus 5 --- README.md | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index c2f1cf6..151922c 100644 --- a/README.md +++ b/README.md @@ -24,10 +24,14 @@

> [!NOTE] -> **Using RampNet exactly as the ICCV'25 paper describes it?** Use the tagged release -> [`v1.0-iccv2025`](https://github.com/ProjectSidewalk/RampNet/releases/tag/v1.0-iccv2025) — the -> repository frozen at paper state. The `main` branch keeps evolving (post-paper analyses, -> benchmarks, and groundwork for what comes next), while the published +> **Looking for a fixed version of this repository?** There are two tagged releases: +> [`v1.0-iccv2025`](https://github.com/ProjectSidewalk/RampNet/releases/tag/v1.0-iccv2025) is the +> repository frozen at paper state — use it to reproduce the ICCV'25 paper exactly as written. +> [`v1.1-corrected-eval`](https://github.com/ProjectSidewalk/RampNet/releases/tag/v1.1-corrected-eval) +> is the same code and the same model weights with the **corrected evaluation protocol** (see the +> [Erratum](#erratum-evaluation-metrics-july-2026) below) — use it to score RampNet's model under +> standard one-to-one matching. The `main` branch keeps evolving (post-paper analyses, benchmarks, +> and groundwork for what comes next), while the published > [model](https://huggingface.co/projectsidewalk/rampnet-model) and > [dataset](https://huggingface.co/datasets/projectsidewalk/rampnet-dataset) on Hugging Face are > versioned independently and match the paper.