Incremental Learning on Food Instance Segmentation

Food instance segmentation is essential to estimate the serving size of dishes in a food image. The recent cutting-edge techniques for instance segmentation are deep learning networks with impressive segmentation quality and fast computation. Nonetheless, they are hungry for data and expensive for annotation. This paper proposes an incremental learning framework to optimize the model performance given a limited data labelling budget. The power of the framework is a novel difficulty assessment model, which forecasts how challenging an unlabelled sample is to the latest trained instance segmentation model. The data collection procedure is divided into several stages, each in which a new sample package is collected. The framework allocates the labelling budget to the most difficult samples. The unlabelled samples that meet a certain qualification from the assessment model are used to generate pseudo-labels. Eventually, the manual labels and pseudo-labels are sent to the training data to improve the instance segmentation model. On four large-scale food datasets, our proposed framework outperforms current incremental learning benchmarks and achieves competitive performance with the model trained on fully annotated samples.

GroundNLQ @ Ego4D Natural Language Queries Challenge 2023

In this report, we present our champion solution for Ego4D Natural Language Queries (NLQ) Challenge in CVPR 2023. Essentially, to accurately ground in a video, an effective egocentric feature extractor and a powerful grounding model are required. Motivated by this, we leverage a two-stage pre-training strategy to train egocentric feature extractors and the grounding model on video narrations, and further fine-tune the model on annotated data. In addition, we introduce a novel grounding model GroundNLQ, which employs a multi-modal multi-scale grounding module for effective video and text fusion and various temporal intervals, especially for long videos. On the blind test set, GroundNLQ achieves 25.67 and 18.18 for R1@IoU=0.3 and R1@IoU=0.5, respectively, and surpasses all other teams by a noticeable margin. Our code will be released at\url{https://github.com/houzhijian/GroundNLQ}.

Interactive Video Corpus Moment Retrieval using Reinforcement Learning

Known-item video search is effective with human-in-the-loop to interactively investigate the search result and refine the initial query. Nevertheless, when the first few pages of results are swamped with visually similar items, or the search target is hidden deep in the ranked list, finding the know-item target usually requires a long duration of browsing and result inspection. This paper tackles the problem by reinforcement learning, aiming to reach a search target within a few rounds of interaction by long-term learning from user feedbacks. Specifically, the system interactively plans for navigation path based on feedback and recommends a potential target that maximizes the long-term reward for user comment. We conduct experiments for the challenging task of video corpus moment retrieval (VCMR) to localize moments from a large video corpus. The experimental results on TVR and DiDeMo datasets verify that our proposed work is effective in retrieving the moments that are hidden deep inside the ranked lists of CONQUER and HERO, which are the state-of-the-art auto-search engines for VCMR.

An Efficient COarse-to-fiNE Alignment Framework @ Ego4D Natural Language Queries Challenge 2022

This technical report describes the CONE approach for Ego4D Natural Language Queries (NLQ) Challenge in ECCV 2022. We leverage our model CONE, an efficient window-centric COarse-to-fiNE alignment framework. Specifically, CONE dynamically slices the long video into candidate windows via a sliding window approach. Centering at windows, CONE (1) learns the inter-window (coarse-grained) semantic variance through contrastive learning and speeds up inference by pre-filtering the candidate windows relevant to the NL query, and (2) conducts intra-window (fine-grained) candidate moments ranking utilizing the powerful multi-modal alignment ability of the contrastive vision-text pre-trained model EgoVLP. On the blind test set, CONE achieves 15.26 and 9.24 for R1@IoU=0.3 and R1@IoU=0.5, respectively.

Dynamic Temporal Filtering in Video Models

Video temporal dynamics is conventionally modeled with 3D spatial-temporal kernel or its factorized version comprised of 2D spatial kernel and 1D temporal kernel. The modeling power, nevertheless, is limited by the fixed window size and static weights of a kernel along the temporal dimension. The pre-determined kernel size severely limits the temporal receptive fields and the fixed weights treat each spatial location across frames equally, resulting in sub-optimal solution for long-range temporal modeling in natural scenes. In this paper, we present a new recipe of temporal feature learning, namely Dynamic Temporal Filter (DTF), that novelly performs spatial-aware temporal modeling in frequency domain with large temporal receptive field. Specifically, DTF dynamically learns a specialized frequency filter for every spatial location to model its long-range temporal dynamics. Meanwhile, the temporal feature of each spatial location is also transformed into frequency feature spectrum via 1D Fast Fourier Transform (FFT). The spectrum is modulated by the learnt frequency filter, and then transformed back to temporal domain with inverse FFT. In addition, to facilitate the learning of frequency filter in DTF, we perform frame-wise aggregation to enhance the primary temporal feature with its temporal neighbors by inter-frame correlation. It is feasible to plug DTF block into ConvNets and Transformer, yielding DTF-Net and DTF-Transformer. Extensive experiments conducted on three datasets demonstrate the superiority of our proposals. More remarkably, DTF-Transformer achieves an accuracy of 83.5% on Kinetics-400 dataset. Source code is available at \url{https://github.com/FuchenUSTC/DTF}.

CONE: An Efficient COarse-to-fiNE Alignment Framework for Long Video Temporal Grounding

Video temporal grounding (VTG) targets to localize temporal moments in an untrimmed video according to a natural language (NL) description. Since real-world applications provide a never-ending video stream, it raises demands for temporal grounding for long-form videos, which leads to two major challenges: (1) the long video length makes it difficult to process the entire video without decreasing sample rate and leads to high computational burden; (2) the accurate multi-modal alignment is more challenging as the number of moment candidates increases. To address these challenges, we propose CONE, an efficient window-centric COarse-to-fiNE alignment framework, which flexibly handles long-form video inputs with higher inference speed, and enhances the temporal grounding via our novel coarse-to-fine multi-modal alignment framework. Specifically, we dynamically slice the long video into candidate windows via a sliding window approach. Centering at windows, CONE (1) learns the inter-window (coarse-grained) semantic variance through contrastive learning and speeds up inference by pre-filtering the candidate windows relevant to the NL query, and (2) conducts intra-window (fine-grained) candidate moments ranking utilizing the powerful multi-modal alignment ability of a contrastive vision-text pre-trained model. Extensive experiments on two large-scale VTG benchmarks for long videos consistently show a substantial performance gain (from 3.13% to 6.87% on MAD and from 10.46% to 13.46% on Ego4d-NLQ) and CONE achieves the SOTA results on both datasets. Analysis reveals the effectiveness of components and higher efficiency in long video grounding as our system improves the inference speed by 2x on Ego4d-NLQ and 15x on MAD while keeping the SOTA performance of CONE.

Long-term Leap Attention, Short-term Periodic Shift for Video Classification

Video transformer naturally incurs a heavier computation burden than a static vision transformer, as the former processes $T$ times longer sequence than the latter under the current attention of quadratic complexity $(T^2N^2)$. The existing works treat the temporal axis as a simple extension of spatial axes, focusing on shortening the spatio-temporal sequence by either generic pooling or local windowing without utilizing temporal redundancy. However, videos naturally contain redundant information between neighboring frames; thereby, we could potentially suppress attention on visually similar frames in a dilated manner. Based on this hypothesis, we propose the LAPS, a long-term ``\textbf{\textit{Leap Attention}}'' (LA), short-term ``\textbf{\textit{Periodic Shift}}'' (\textit{P}-Shift) module for video transformers, with $(2TN^2)$ complexity. Specifically, the ``LA'' groups long-term frames into pairs, then refactors each discrete pair via attention. The ``\textit{P}-Shift'' exchanges features between temporal neighbors to confront the loss of short-term dynamics. By replacing a vanilla 2D attention with the LAPS, we could adapt a static transformer into a video one, with zero extra parameters and neglectable computation overhead ($\sim$2.6\%). Experiments on the standard Kinetics-400 benchmark demonstrate that our LAPS transformer could achieve competitive performances in terms of accuracy, FLOPs, and Params among CNN and transformer SOTAs. We open-source our project in \sloppy \href{https://github.com/VideoNetworks/LAPS-transformer}{\textit{\color{magenta}{https://github.com/VideoNetworks/LAPS-transformer}}} .

Wave-ViT: Unifying Wavelet and Transformers for Visual Representation Learning

Multi-scale Vision Transformer (ViT) has emerged as a powerful backbone for computer vision tasks, while the self-attention computation in Transformer scales quadratically w.r.t. the input patch number. Thus, existing solutions commonly employ down-sampling operations (e.g., average pooling) over keys/values to dramatically reduce the computational cost. In this work, we argue that such over-aggressive down-sampling design is not invertible and inevitably causes information dropping especially for high-frequency components in objects (e.g., texture details). Motivated by the wavelet theory, we construct a new Wavelet Vision Transformer (\textbf{Wave-ViT}) that formulates the invertible down-sampling with wavelet transforms and self-attention learning in a unified way. This proposal enables self-attention learning with lossless down-sampling over keys/values, facilitating the pursuing of a better efficiency-vs-accuracy trade-off. Furthermore, inverse wavelet transforms are leveraged to strengthen self-attention outputs by aggregating local contexts with enlarged receptive field. We validate the superiority of Wave-ViT through extensive experiments over multiple vision tasks (e.g., image recognition, object detection and instance segmentation). Its performances surpass state-of-the-art ViT backbones with comparable FLOPs. Source code is available at \url{https://github.com/YehLi/ImageNetModel}.

(Un)likelihood Training for Interpretable Embedding

Cross-modal representation learning has become a new normal for bridging the semantic gap between text and visual data. Learning modality agnostic representations in a continuous latent space, however, is often treated as a black-box data-driven training process. It is well-known that the effectiveness of representation learning depends heavily on the quality and scale of training data. For video representation learning, having a complete set of labels that annotate the full spectrum of video content for training is highly difficult if not impossible. These issues, black-box training and dataset bias, make representation learning practically challenging to be deployed for video understanding due to unexplainable and unpredictable results. In this paper, we propose two novel training objectives, likelihood and unlikelihood functions, to unroll semantics behind embeddings while addressing the label sparsity problem in training. The likelihood training aims to interpret semantics of embeddings beyond training labels, while the unlikelihood training leverages prior knowledge for regularization to ensure semantically coherent interpretation. With both training objectives, a new encoder-decoder network, which learns interpretable cross-modal representation, is proposed for ad-hoc video search. Extensive experiments on TRECVid and MSR-VTT datasets show the proposed network outperforms several state-of-the-art retrieval models with a statistically significant performance margin.

MLP-3D: A MLP-like 3D Architecture with Grouped Time Mixing

Convolutional Neural Networks (CNNs) have been regarded as the go-to models for visual recognition. More recently, convolution-free networks, based on multi-head self-attention (MSA) or multi-layer perceptrons (MLPs), become more and more popular. Nevertheless, it is not trivial when utilizing these newly-minted networks for video recognition due to the large variations and complexities in video data. In this paper, we present MLP-3D networks, a novel MLP-like 3D architecture for video recognition. Specifically, the architecture consists of MLP-3D blocks, where each block contains one MLP applied across tokens (i.e., token-mixing MLP) and one MLP applied independently to each token (i.e., channel MLP). By deriving the novel grouped time mixing (GTM) operations, we equip the basic token-mixing MLP with the ability of temporal modeling. GTM divides the input tokens into several temporal groups and linearly maps the tokens in each group with the shared projection matrix. Furthermore, we devise several variants of GTM with different grouping strategies, and compose each variant in different blocks of MLP-3D network by greedy architecture search. Without the dependence on convolutions or attention mechanisms, our MLP-3D networks achieves 68.5\%/81.4\% top-1 accuracy on Something-Something V2 and Kinetics-400 datasets, respectively. Despite with fewer computations, the results are comparable to state-of-the-art widely-used 3D CNNs and video transformers. Source code is available at https://github.com/ZhaofanQiu/MLP-3D.