ترغب بنشر مسار تعليمي؟ اضغط هنا

Less is More: Pay Less Attention in Vision Transformers

82   0   0.0 ( 0 )
 نشر من قبل Bohan Zhuang
 تاريخ النشر 2021
  مجال البحث الهندسة المعلوماتية
والبحث باللغة English




اسأل ChatGPT حول البحث

Transformers have become one of the dominant architectures in deep learning, particularly as a powerful alternative to convolutional neural networks (CNNs) in computer vision. However, Transformer training and inference in previous works can be prohibitively expensive due to the quadratic complexity of self-attention over a long sequence of representations, especially for high-resolution dense prediction tasks. To this end, we present a novel Less attention vIsion Transformer (LIT), building upon the fact that convolutions, fully-connected (FC) layers, and self-attentions have almost equivalent mathematical expressions for processing image patch sequences. Specifically, we propose a hierarchical Transformer where we use pure multi-layer perceptrons (MLPs) to encode rich local patterns in the early stages while applying self-attention modules to capture longer dependencies in deeper layers. Moreover, we further propose a learned deformable token merging module to adaptively fuse informative patches in a non-uniform manner. The proposed LIT achieves promising performance on image recognition tasks, including image classification, object detection and instance segmentation, serving as a strong backbone for many vision tasks. Code is available at: https://github.com/MonashAI/LIT



قيم البحث

اقرأ أيضاً

Self-attention is a useful mechanism to build generative models for language and images. It determines the importance of context elements by comparing each element to the current time step. In this paper, we show that a very lightweight convolution c an perform competitively to the best reported self-attention results. Next, we introduce dynamic convolutions which are simpler and more efficient than self-attention. We predict separate convolution kernels based solely on the current time-step in order to determine the importance of context elements. The number of operations required by this approach scales linearly in the input length, whereas self-attention is quadratic. Experiments on large-scale machine translation, language modeling and abstractive summarization show that dynamic convolutions improve over strong self-attention models. On the WMT14 English-German test set dynamic convolutions achieve a new state of the art of 29.7 BLEU.
Highlight detection has the potential to significantly ease video browsing, but existing methods often suffer from expensive supervision requirements, where human viewers must manually identify highlights in training videos. We propose a scalable uns upervised solution that exploits video duration as an implicit supervision signal. Our key insight is that video segments from shorter user-generated videos are more likely to be highlights than those from longer videos, since users tend to be more selective about the content when capturing shorter videos. Leveraging this insight, we introduce a novel ranking framework that prefers segments from shorter videos, while properly accounting for the inherent noise in the (unlabeled) training data. We use it to train a highlight detector with 10M hashtagged Instagram videos. In experiments on two challenging public video highlight detection benchmarks, our method substantially improves the state-of-the-art for unsupervised highlight detection.
Obtaining viewer responses from videos can be useful for creators and streaming platforms to analyze the video performance and improve the future user experience. In this report, we present our method for 2021 Evoked Expression from Videos Challenge. In particular, our model utilizes both audio and image modalities as inputs to predict emotion changes of viewers. To model long-range emotion changes, we use a GRU-based model to predict one sparse signal with 1Hz. We observe that the emotion changes are smooth. Therefore, the final dense prediction is obtained via linear interpolating the signal, which is robust to the prediction fluctuation. Albeit simple, the proposed method has achieved pearsons correlation score of 0.04430 on the final private test set.
We investigate the steady state of a system of photons in a pumped dye-filled microcavity. By varying pump and thermalization the system can be tuned between Bose-Einstein condensation, multimode condensation, and lasing. We present a rich non-equili brium phase diagram which exhibits transitions between these phases, including decondensation of individual modes under conditions that would typically favor condensation.
An abdominal ultrasound examination, which is the most common ultrasound examination, requires substantial manual efforts to acquire standard abdominal organ views, annotate the views in texts, and record clinically relevant organ measurements. Hence , automatic view classification and landmark detection of the organs can be instrumental to streamline the examination workflow. However, this is a challenging problem given not only the inherent difficulties from the ultrasound modality, e.g., low contrast and large variations, but also the heterogeneity across tasks, i.e., one classification task for all views, and then one landmark detection task for each relevant view. While convolutional neural networks (CNN) have demonstrated more promising outcomes on ultrasound image analytics than traditional machine learning approaches, it becomes impractical to deploy multiple networks (one for each task) due to the limited computational and memory resources on most existing ultrasound scanners. To overcome such limits, we propose a multi-task learning framework to handle all the tasks by a single network. This network is integrated to perform view classification and landmark detection simultaneously; it is also equipped with global convolutional kernels, coordinate constraints, and a conditional adversarial module to leverage the performances. In an experimental study based on 187,219 ultrasound images, with the proposed simplified approach we achieve (1) view classification accuracy better than the agreement between two clinical experts and (2) landmark-based measurement errors on par with inter-user variability. The multi-task approach also benefits from sharing the feature extraction during the training process across all tasks and, as a result, outperforms the approaches that address each task individually.
التعليقات
جاري جلب التعليقات جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
mircosoft-partner

هل ترغب بارسال اشعارات عن اخر التحديثات في شمرا-اكاديميا