Quo vadis?

By Тараса Шевченка, Василя Каразіна, Олександра Богомольця, О. О. Богомольця. 2013

822 citations recorded by the source corpus.

How researchers see this paper

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

In Philosophy

Citations in the corpus

Rethinking ImageNet Pre-Training (2018, background): “Pre-training has enabled state-of-the-art results on many tasks, including object detection [9, 8, 36], image segmentation [29, 13], and action recognition [42, 4].”
Group Normalization (2018, methodology): “For example, the Fast/er and Mask R-CNN frameworks [12, 47, 18] use a batch size of 1 or 2 images because of higher resolution, where BN is “frozen” by transforming to a linear layer [20]; in video classification with 3D convolutions [60, 6], the presence of spatial-temporal features introduces a tr”
Self-Supervised MultiModal Versatile Networks (2020, methodology): “In the two types of video networks considered here, this deflation corresponds to the following operations: for 3D convolutional based networks [15, 90], summing the 3D spatiotemporal filters over the temporal dimension to obtain 2D filters; for TSM networks [47], turning off the channel shifting wh”
A Short Note on the Kinetics-700 Human Action Dataset (2019, methodology): “The top team was from JD AI Research and obtained 17.9% error, considerably below our baseline – a single RGB I3D model – which obtained 29.3% error.”
HowTo100M: Learning a Text-Video Embedding by Watching Hundred Million Narrated Video Clips (2019, methodology): “3D features are extracted with the Kinetics [4] pre-trained ResNeXt-101 16-frames model [12] to obtain 1.5 features per second.”
A Short Note on the Kinetics-700-2020 Human Action Dataset (2020, methodology): “Top-1 and top-5 accuracy improve steadily with more examples per class, even given that I3D is a model with few parameters: around 12M.”
AVA: A Video Dataset of Spatio-Temporally Localized Atomic Visual Actions (2017, methodology): “The I3D architecture is designed based on the Inception architecture [40], but replaces 2D convolutions with 3D convo-lutions.”
AssembleNet: Searching for Multi-Stream Neural Connectivity in Video Architectures (2019, neutral): “For instance, another large-scale video dataset, Kinetics [3], is not a static dataset.”
A Closer Look at Spatiotemporal Convolutions for Action Recognition (2017, methodology): “We train our R(2+1)D architecture on both RGB and optical flow inputs and fuse the prediction scores by averaging, as proposed in the two-stream framework [29] and subsequent work [4, 9].”
A Short Note about Kinetics-600 (2018, methodology): “As a baseline model we used I3D [2], with standard RGB videos as input (no optical flow).”
Temporal Relational Reasoning in Videos (2017, background): “Thus, activity recognition networks such as Two Stream network [23] and the I3D network [3] are tailored to capture these short-term dynamics of dense frames.”
Video Action Transformer Network (2018, methodology): “R50: ResNet-50 [17], I3D: Inflated 3D convolutions [7], S3D(+G): Separable 3D convolutions (with gating) [48], FRCNN: Faster RCNN [33], NL: Non-local networks [46], P3D: Pseudo-3D convolutions [32], C2D [42], C3D [42], TSN: Temporal Segment Networks [45] RN: Relation Nets [34], Tx: Transformer [31, ”
Video-to-Video Synthesis (2018, methodology): “We use two different pre-trained video recognition CNNs in our evaluation: I3D [8] and ResNeXt [76].”
Long-Term Feature Banks for Detailed Video Understanding (2018, methodology): “, C3D [46], I3D [5]) with a region-based person detector (e.”
Unidentified Video Objects: A Benchmark for Dense, Open-World Segmentation (2021, background): “Current video understanding approaches [36, 5, 38, 12, 37] cannot scale well to long videos due to GPU memory constraints and are not designed for complex prediction tasks beyond classification and detection.”
From Lifestyle Vlogs to Everyday Interactions (2017, methodology): “We train a linear SVM on average activations (I3D-K); as well as fine-tune it on VLOG (FT I3D-K).”
Deep Video Inpainting (2019, methodology): “As in [26], we follow the protocol that uses the I3D network [5] pretrained on a video recognition task to measure the distance between the spatiotemporal features extracted from the output videos and the ground-truth videos.”
Progressive Teacher-Student Learning for Early Action Prediction (2019, background): “Recently, with the rise of deep learning, many deep learning based methods, including 3D CNN [39] [5] [12] [40] [13] and two stream CNN [36] [8] [46] etc., are proposed to encode the spatiotemporal information and achieve satisfactory recognition results in many datasets, including UCF-101 [37] and ”