Conventionally, spatiotemporal modeling network and its complexity are the two most concentrated research topics in video action recognition. Existing state-of-the-art methods have achieved excellent accuracy regardless of the complexity meanwhile efficient spatiotemporal modeling solutions are slightly inferior in performance. In this paper, we attempt to acquire both efficiency and effectiveness simultaneously. First of all, besides traditionally treating H x W x T video frames as space-time signal (viewing from the Height-Width spatial plane), we propose to also model video from the other two Height-Time and Width-Time planes, to capture the dynamics of video thoroughly. Secondly, our model is designed based on 2D CNN backbones and model complexity is well kept in mind by design. Specifically, we introduce a novel multi-view fusion (MVF) module to exploit video dynamics using separable convolution for efficiency. It is a plug-and-play module and can be inserted into off-the-shelf 2D CNNs to form a simple yet effective model called MVFNet. Moreover, MVFNet can be thought of as a generalized video modeling framework and it can specialize to be existing methods such as C2D, SlowOnly, and TSM under different settings. Extensive experiments are conducted on popular benchmarks (i.e., Something-Something V1 & V2, Kinetics, UCF-101, and HMDB-51) to show its superiority. The proposed MVFNet can achieve state-of-the-art performance with 2D CNN's complexity.
Recovering 3D human mesh from monocular images is a popular topic in computer vision and has a wide range of applications. This paper aims to estimate 3D mesh of multiple body parts (e.g., body, hands) with large-scale differences from a single RGB image. Existing methods are mostly based on iterative optimization, which is very time-consuming. We propose to train a single-shot model to achieve this goal. The main challenge is lacking training data that have complete 3D annotations of all body parts in 2D images. To solve this problem, we design a multi-branch framework to disentangle the regression of different body properties, enabling us to separate each component's training in a synthetic training manner using unpaired data available. Besides, to strengthen the generalization ability, most existing methods have used in-the-wild 2D pose datasets to supervise the estimated 3D pose via 3D-to-2D projection. However, we observe that the commonly used weak-perspective model performs poorly in dealing with the external foreshortening effect of camera projection. Therefore, we propose a depth-to-scale (D2S) projection to incorporate the depth difference into the projection function to derive per-joint scale variants for more proper supervision. The proposed method outperforms previous methods on the CMU Panoptic Studio dataset according to the evaluation results and achieves comparable results on the Human3.6M body and STB hand benchmarks. More impressively, the performance in close shot images gets significantly improved using the proposed D2S projection for weak supervision, while maintains obvious superiority in computational efficiency.
Interactive Fiction (IF) games with real human-written natural language texts provide a new natural evaluation for language understanding techniques. In contrast to previous text games with mostly synthetic texts, IF games pose language understanding challenges on the human-written textual descriptions of diverse and sophisticated game worlds and language generation challenges on the action command generation from less restricted combinatorial space. We take a novel perspective of IF game solving and re-formulate it as Multi-Passage Reading Comprehension (MPRC) tasks. Our approaches utilize the context-query attention mechanisms and the structured prediction in MPRC to efficiently generate and evaluate action outputs and apply an object-centric historical observation retrieval strategy to mitigate the partial observability of the textual observations. Extensive experiments on the recent IF benchmark (Jericho) demonstrate clear advantages of our approaches achieving high winning rates and low data requirements compared to all previous approaches. Our source code is available at: https://github.com/XiaoxiaoGuo/rcdqn.
We addressed the challenging task of video question answering, which requires machines to answer questions about videos in a natural language form. Previous state-of-the-art methods attempt to apply spatio-temporal attention mechanism on video frame features without explicitly modeling the location and relations among object interaction occurred in videos. However, the relations between object interaction and their location information are very critical for both action recognition and question reasoning. In this work, we propose to represent the contents in the video as a location-aware graph by incorporating the location information of an object into the graph construction. Here, each node is associated with an object represented by its appearance and location features. Based on the constructed graph, we propose to use graph convolution to infer both the category and temporal locations of an action. As the graph is built on objects, our method is able to focus on the foreground action contents for better video question answering. Lastly, we leverage an attention mechanism to combine the output of graph convolution and encoded question features for final answer reasoning. Extensive experiments demonstrate the effectiveness of the proposed methods. Specifically, our method significantly outperforms state-of-the-art methods on TGIF-QA, Youtube2Text-QA, and MSVD-QA datasets. Code and pre-trained models are publicly available at: https://github.com/SunDoge/L-GCN
We present Tiny-Transfer-Learning (TinyTL), an efficient on-device learning method to adapt pre-trained models to newly collected data on edge devices. Different from conventional transfer learning methods that fine-tune the full network or the last layer, TinyTL freezes the weights of the feature extractor while only learning the biases, thus doesn't require storing the intermediate activations, which is the major memory bottleneck for on-device learning. To maintain the adaptation capacity without updating the weights, TinyTL introduces memory-efficient lite residual modules to refine the feature extractor by learning small residual feature maps in the middle. Besides, instead of using the same feature extractor, TinyTL adapts the architecture of the feature extractor to fit different target datasets while fixing the weights: TinyTL pre-trains a large super-net that contains many weight-shared sub-nets that can individually operate; different target dataset selects the sub-net that best match the dataset. This backpropagation-free discrete sub-net selection incurs no memory overhead. Extensive experiments show that TinyTL can reduce the training memory cost by order of magnitude (up to 13.3x) without sacrificing accuracy compared to fine-tuning the full network.
Humans integrate multiple sensory modalities (e.g. visual and audio) to build a causal understanding of the physical world. In this work, we propose a novel type of intrinsic motivation for Reinforcement Learning (RL) that encourages the agent to understand the causal effect of its actions through auditory event prediction. First, we allow the agent to collect a small amount of acoustic data and use K-means to discover underlying auditory event clusters. We then train a neural network to predict the auditory events and use the prediction errors as intrinsic rewards to guide RL exploration. Experimental results on Atari games show that our new intrinsic motivation significantly outperforms several state-of-the-art baselines. We further visualize our noisy agents' behavior in a physics environment and demonstrate that our newly designed intrinsic reward leads to the emergence of physical interaction behaviors (e.g. contact with objects).
In this paper, we introduce Foley Music, a system that can synthesize plausible music for a silent video clip about people playing musical instruments. We first identify two key intermediate representations for a successful video to music generator: body keypoints from videos and MIDI events from audio recordings. We then formulate music generation from videos as a motion-to-MIDI translation problem. We present a Graph$-$Transformer framework that can accurately predict MIDI event sequences in accordance with the body movements. The MIDI event can then be converted to realistic music using an off-the-shelf music synthesizer tool. We demonstrate the effectiveness of our models on videos containing a variety of music performances. Experimental results show that our model outperforms several existing systems in generating music that is pleasant to listen to. More importantly, the MIDI representations are fully interpretable and transparent, thus enabling us to perform music editing flexibly. We encourage the readers to watch the demo video with audio turned on to experience the results.
Machine learning on tiny IoT devices based on microcontroller units (MCU) is appealing but challenging: the memory of microcontrollers is 2-3 orders of magnitude less even than mobile phones. We propose MCUNet, a framework that jointly designs the efficient neural architecture (TinyNAS) and the lightweight inference engine (TinyEngine), enabling ImageNet-scale inference on microcontrollers. TinyNAS adopts a two-stage neural architecture search approach that first optimizes the search space to fit the resource constraints, then specializes the network architecture in the optimized search space. TinyNAS can automatically handle diverse constraints (i.e. device, latency, energy, memory) under low search costs. TinyNAS is co-designed with TinyEngine, a memory-efficient inference library to expand the design space and fit a larger model. TinyEngine adapts the memory scheduling according to the overall network topology rather than layer-wise optimization, reducing the memory usage by 2.7x, and accelerating the inference by 1.7-3.3x compared to TF-Lite Micro and CMSIS-NN. MCUNet is the first to achieves >70% ImageNet top1 accuracy on an off-the-shelf commercial microcontroller, using 3.6x less SRAM and 6.6x less Flash compared to quantized MobileNetV2 and ResNet-18. On visual&audio wake words tasks, MCUNet achieves state-of-the-art accuracy and runs 2.4-3.4x faster than MobileNetV2 and ProxylessNAS-based solutions with 2.2-2.6x smaller peak SRAM. Our study suggests that the era of always-on tiny machine learning on IoT devices has arrived.
We focus on the task of generating sound from natural videos, and the sound should be both temporally and content-wise aligned with visual signals. This task is extremely challenging because some sounds generated \emph{outside} a camera can not be inferred from video content. The model may be forced to learn an incorrect mapping between visual content and these irrelevant sounds. To address this challenge, we propose a framework named REGNET. In this framework, we first extract appearance and motion features from video frames to better distinguish the object that emits sound from complex background information. We then introduce an innovative audio forwarding regularizer that directly considers the real sound as input and outputs bottlenecked sound features. Using both visual and bottlenecked sound features for sound prediction during training provides stronger supervision for the sound prediction. The audio forwarding regularizer can control the irrelevant sound component and thus prevent the model from learning an incorrect mapping between video frames and sound emitted by the object that is out of the screen. During testing, the audio forwarding regularizer is removed to ensure that REGNET can produce purely aligned sound only from visual features. Extensive evaluations based on Amazon Mechanical Turk demonstrate that our method significantly improves both temporal and content-wise alignment. Remarkably, our generated sound can fool the human with a 68.12% success rate. Code and pre-trained models are publicly available at https://github.com/PeihaoChen/regnet