The EgoProceL dataset was released in an ECCV 2022 publication titled: "My View is the Best View: Procedure Learning from Egocentric Videos".
Link to download EgoProceL: https://sid2697.github.io/egoprocel/#download
Link to the project page: https://sid2697.github.io/egoprocel/
Link to the paper: Coming soon!
This document summarizes the steps required to download the videos and annotations in EgoProceL.
Link to download: OneDrive (expired) Google Drive (please use this)
Folder structure:
+ annotations
+ CMU-MMAC
+ Brownie
+ S07_Brownie_Video
- S07_Brownie_6510211-1103.csv
- S07_Brownie_7150991-1431.csv
- S07_Brownie_7151020-1103.csv
- S07_Brownie_7151062-1103.csv
- S07_Brownie_8421130-2374.csv
...
+ Eggs
+ S07_Eggs_Video
- S07_Eggs_6510211-1110.csv
...
...
...
+ EGTEA_Gaze+
+ BaconAndEggs
- OP01-R03-BaconAndEggs.csv
- OP02-R03-BaconAndEggs.csv
...
+ Cheeseburger
...
...
+ EPIC-Tents
- 01.tent.090617.gopro.egoprocel.ann.csv
- 02.tent.120617.gopro.egoprocel.ann.csv
...
+ MECCANO
- 0003.csv
- 0004.csv
...
+ pc_assembly
- Head_5.csv
- Head_7.csv
...
+ pc_disassembly
- Head_6.csv
- Head_8.csv
...
Things to note about the annotations:
- The annotation file name exactly matches the video's file name.
- The annotation
csv
contains three columns, a) key-step's start second, b) key-step's end second, c) name of the key-step. - The datasets with multiple categories (e.g., CMU-MMAC) have multiple directories under them. In contrast, datasets with single category (e.g., PC Assembly) directly have the annotation
csv
.
The videos in EgoProceL were obtained from multiple sources. Here we list the steps to download the videos from each of the sources:
These videos were recorded by Pravin Nagar and Sagar Verma at IIIT Delhi.
Download link: OneDrive Google Drive (please use this)
CMU-MMAC videos can be downloaded from http://kitchen.cs.cmu.edu/main.php.
Here is a script to download all the videos at once: https://github.com/Sid2697/EgoProceL-egocentric-procedure-learning/blob/main/misc/CMU_Kitchens/download.py.
EGTEA-Gaze+ videos can be downloaded from https://cbs.ic.gatech.edu/fpv/.
EPIC-Tents videos can be downloaded from https://sites.google.com/view/epic-tent.
MECCANO videos can be downloaded from https://iplab.dmi.unict.it/MECCANO/.
Things to note about the videos:
- It is recommended to save the videos following the annotation directory's structure.
- Due to compatibility reasons (mentioned in the paper), not all the videos from each dataset have been used for the task. Please refer to the available annotation files to get an idea of which videos to use.
In case of any concern contact Siddhant Bansal. Email: siddhant.bansal@research.iiit.ac.in
Please consider citing if you make use of the EgoProceL dataset and/or the corresponding code:
@InProceedings{EgoProceLECCV2022,
author="Bansal, Siddhant
and Arora, Chetan
and Jawahar, C.V.",
title="My View is the Best View: Procedure Learning from Egocentric Videos",
booktitle = "European Conference on Computer Vision (ECCV)",
year="2022"
}