DRIV100 (Diverse Roadscenes from Internet Videos 100) ================== Created By ---------- Haruya Sakashita*, Christoph Flothow^, Noriko Takemura*, Yusuke Sugano† * Osaka University ^ Technische Universität Darmstadt † The University of Tokyo Version 1.0 Description ----------- The DRIV100 dataset consists of 100 YouTube videos selected with Human-Judgment/Anomaly-Score based criteria to guarantee the scene diversity. The following technical report contains a detailed description of the dataset and how it was compiled: H. Sakashita, C. Flothow, N. Takemura and Y. Sugano, "In-The-Wild Multi-Domain Dataset and Evaluation for Real-World Domain Adaptation of Semantic Segmentation", https://arxiv.org/abs/2102.00150 Files Included -------------------- - json This directory contains the label data (JSON files) for the evaluation frame. - meta.csv This file contains the YouTube ids ("URL"), annotation category ("category"), evaluation frame ("eval_frame"), start frame ("start_frame"), and end frame ("end_frame") of the videos in the DRIV100. - json2label.py This sample script generates colored label images or binary index label images from the JSON label data. - frame_extract.py This sample script extracts the frames for domain adaptation. This script requires 720p Youtube videos of DRIV100 stored in the "DRIV100" directory. It saves the 2,329 frames used in our technical report, together with frame lists for experiments using LSD and CRST. With the "--output-test-frames" option, it saves the 400 test frames of DRIV100. Please Acknowledge DRIV100 in Academic Research ---------------------------------------------------- When DRIV100 is used for academic research, we would highly appreciate it if scientific publications of works cite the following technical report: H. Sakashita, C. Flothow, N. Takemura and Y. Sugano, "In-The-Wild Multi-Domain Dataset and Evaluation for Real-World Domain Adaptation of Semantic Segmentation", https://arxiv.org/abs/2102.00150 Conditions of Use ----------------- Dataset compiled by Haruya Sakashita, Christoph Flothow, Noriko Takemura and Yusuke Sugano. The dataset and its contents are made available on an "as is" basis and without warranties of any kind, including without limitation satisfactory quality and conformity, merchantability, fitness for a particular purpose, accuracy or completeness, or absence of errors. Subject to any liability that may not be excluded or limited by law, Osaka University, Technische Universität Darmstadt and The University of Tokyo are not liable for, and expressly excludes, all liability for loss or damage however and whenever caused to anyone by any use of the DRIV100 dataset or any part of it. DRIV100 is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.