Large scale activity recognition has made remarkable progress driven by the curation of large scale labeled video datasets. Evaluation tasks in these datasets include activity classification, activity and object detection and localization, action prediction, episodic memory for object instance retrieval, object interactions with hands/tools/people, speaker prediction and scene diarization in long duration videos.
However, performance on these important tasks remains limited by the scale, quality and applicability of data. While there are many large-scale video datasets for pretraining activity recognition, these datasets are nearly all scraped from social media platforms such as YouTube or Instagram. The activities in these datasets do not reflect true diversity of activities in the real world, and instead represent a biased and unbalanced sample of labels that are interesting enough for social media. Specifically, these datasets lack coverage of:
Our goal is to bring together researchers interested in the problem of activity recognition of common, fine and diverse human activities. We will provide a forum for discussion, and provide a new large-scale training/validation/test set of fine-grained activities collected from around the world to focus discussion and characterize progress in this emerging research field.
We will have invited keynote talks from experts in the field. We will also have two talks from the top performers on the Open Fine-grained Activity Detection (OpenFAD) Activity Classification and Temporal Activity Detection challenges that uses the Consented Activities of People (CAP) dataset.
A half day workshop on Fine Grained Activity Detection to be held on October 2, 2023 at the 2nd Workshop on Fine Grained Activity Detection to be held on October 2, 2023 at the ICCV’23 in Paris, France.
Every day, humans perform many closely related activities that involve subtle discriminative motions, such as putting on a shirt vs. putting on a jacket, or shaking hands vs. giving a high five. Activity recognition by ethical visual AI could provide insights into our patterns of daily life, however existing activity recognition datasets do not capture the massive diversity of these human activities around the world. The goal of this workshop is to bring together research the representation and recognition of fine grained activities of daily life around the world, and provide a forum for researchers to share expertise on this emerging problem.
The workshop will host the second Open Fine-grained Activity Detection (OpenFAD) Challenge. OpenFAD is an activity classification and detection evaluation to measure how well systems can automatically classify or temporally detect fine-grained activities collected from handheld devices. This challenge uses the Consented Activities of People (CAP) dataset which is a new dataset of 1.45M videos of 512 fine grained activity classes of consented people. We will host a leaderboard for the Activity Classification (AC) and Temporal Activity Detection (TAD) submission tracks to characterize the state-of-the-art on this new dataset. The leaderboard evaluation will be performed on a video test set available to the challenge participants, with sequestered ground truth.
Accepted papers will be published in the workshop proceedings, and the top two teams on the leaderboard will be invited to give a talk.
|May 5, 2023||TAD/AC evaluation leaderboard opens|
| September 15
||TAD/AC submission deadline|
| September 18
||Challenge Winner notification|
| September 21
||TAD/AC final report available|
|October 2, 2023||2nd Fine Grained Activity Detection (ICCV FGAD'23) Workshop at ICCV23|
|09:00 - 09:05:||Welcome and introduction|
|09:05 - 09:30:||Jeffrey Byrne, Visym Labs, Fine Grained Activity Detection (FGAD'23) Challenge Overview|
|09:30 - 10:15:||Limin Wang, Nanjing University, Invited Talk #1 (45 min)|
|10:15 - 10:45:||Jon Fiscus, NIST, Fine Grained Activity Detection (FGAD'23) Challenge Results|
|10:45 - 11:00:||Coffee Break|
|11:00 - 11:20:||Activity Classification Challenge Co-Winner, DeepGlint|
|11:20 - 11:40:||Activity Classification Challenge Co-Winner, Cloudwalk Technology|
|11:40 - 12:00:||Temporal Activity Detection Challenge Winner, WasedaMeiseiSoftbank|
|12:00 - 12:45:||Joao Carreria, Google DeepMind, Invited Talk #2 (45 min)|
|12:45 - 13:00:||Open discussion|
Participants are encouraged to submit results to the challenge leaderboard. The workshop will feature an in-person poster session for participants to share the design of submitted systems. Our goal is to bring together researchers interested in the problem of activity recognition of common, fine and diverse human activities, and support discussion on this emerging topic.
You are able to register for the ICCV FGAD'23 workshop by attending ICCV'23.