RESOUND: Towards Action Recognition Without Representation Bias

By Yingwei Li, Nuno Vasconcelos. 2018

415 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In AI Ethics, Fairness & Society

Citations in the corpus

MERLOT: Multimodal Neural Script Knowledge Models (2021, background): “However, common benchmarks for activity recognition tend to have strong object and background bias – for example, to recognize the UCF action ‘playing guitar,’ it is sufﬁcient to detect a guitar in an image (as guitars are unlikely to show up for the other activities like ‘playing basketball’) [70].”
Adversarial Filters of Dataset Biases (2020, background): “RESOUND (Li et al., 2018) quantifies the representation biases of datasets.”
Unbiased Scene Graph Generation From Biased Training (2020, neutral): “augmentation or re-sampling [9, 25, 27, 11, 3], 2) unbiased learning through elaborately designed training curriculums or learning losses [70, 30], 3) disentangling biased representations from the unbiased [35, 4].”
CATER: A diagnostic dataset for Compositional Actions and TEmporal Reasoning (2019, background): “While some recent work (Goyal et al., 2017; Li et al., 2018) attempts to control for this bias, it still remains a challenge for long-term reasoning tasks.”
FineGym: A Hierarchical Video Dataset for Fine-Grained Action Understanding (2020, background): “There are also attempts towards building datasets for fine-grained action recognition [6, 34, 19, 15, 24, 29].”
Only Time Can Tell: Discovering Temporal Data for Temporal Modeling (2019, background): “VLOG [10]), action sequence classification [23], and video prediction (e.”
Learning De-biased Representations with Biased Representations (2019, methodology): “(1) The ﬁrst is a data re-weighting solution: we give greater weights on frogs in sky than frogs in swamps to even out the correlation in p ( B, Y ) (Li et al., 2018; Li & Vasconcelos, 2019).”
VIMPAC: Video Pre-Training via Masked Token Prediction and Contrastive Learning (2021, methodology): “For downstream evaluation, we experiment with several action classification datasets: UCF101 [56], HMDB51 [37], Kinetics-400 [9], SSV2 [26], and Diving48 [41].”
MARS: Motion-Augmented RGB Stream for Action Recognition (2019, background): “[22] also ﬁnd that Kinetics400 is biased towards static information.”
Coherent Temporal Synthesis for Incremental Action Segmentation (2024, neutral): “Also, there is a strong static bias towards scene context [30] in standard action recognition datasets like Ki-netics [7] and UCF101 [46].”
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (2025, neutral): “…of the model’s learned representation on a set of six motion and appearance classification tasks: Something-Something v2 (Goyal et al., 2017), Diving-48 (Li et al., 2018), Jester (Materzynska et al., 2019), Kinetics (Kay et al., 2017), COIN (Tang et al., 2019), and ImageNet (Deng et al., 2009).”
Grouped Spatial-Temporal Aggregation for Efficient Action Recognition (2019, methodology): “Diving48 Diving48 [13] is a newly released dataset with more than 18K video clips for 48 diving classes.”
REPAIR: Removing Representation Bias by Dataset Resampling (2019, background): “It was ﬁrst explicitly formulated in [22], and used to measure the bias of modern action recognition datasets towards objects, scenes and people.”
Why Can't I Dance in the Mall? Learning to Mitigate Scene Bias in Action Recognition (2019, methodology): “For the action classiﬁcation task, we evaluate the transfer learning performance on the UCF-101 [50], HMDB-51 [34], and Diving48 [38] datasets.”
BEVT: BERT Pretraining of Video Transformers (2021, methodology): “On the other hand, temporal dynamics are crucial for differentiating actions be-tween two ﬁne-grained diving sequences [33].”