Feedback Networks

Zamir, Amir; Wu, Te-Lin; Sun, Lin; Shen, William; Shi, Bertram E.; Malik, Jitendra; Savarese, Silvio

doi:10.1109/cvpr.2017.196

Cited by 148 publications

(110 citation statements)

References 32 publications

Supporting

Mentioning

109

Contrasting

Order By: Relevance

“…We used Adam Optimizer with learning rate l r = 1e − 4 during 50 epochs, selecting such a small learning rate was crucial for training the BLSTM for avoiding the well-know gradient vanishing problem on Recurrent Neural Networks. Both temporal filters boosted the performance on the Precision (mAP on T-LESS and P on CORe50); related to the BLSTM, the results are consistent on works like in [20], in which the performance gets better using more cells and with the number of neurons being not as relevant as the number of cells.…”

Section: Temporal Componentsupporting

confidence: 72%

Instance-level Object Recognition Using Deep Temporal Coherence

Lagunes-Fortiz

Damen

Mayol-Cuevas

2018

Advances in Visual Computing

View full text Add to dashboard Cite

In this paper we design and evaluate methods for exploiting temporal coherence present in video data for the task of instance object recognition. First, we evaluate the performance and generalisation capabilities of a Convolutional Neural Network for learning individual objects from multiple viewpoints coming from a video sequence. Then, we exploit the assumption that on video data the same object remains present over a number of consecutive frames. A-priori knowing such number of consecutive frames is a difficult task however, specially for mobile agents interacting with objects in front of them. Thus, we evaluate the use of temporal filters such as Cumulative Moving Average and a machine learning approach using Recurrent Neural Networks for this task. We also show that by exploiting temporal coherence, models trained with a few data points perform comparably to when the whole dataset is available.

show abstract

Section: Temporal Componentsupporting

confidence: 72%

Instance-level Object Recognition Using Deep Temporal Coherence

Lagunes-Fortiz

Damen

Mayol-Cuevas

2018

Advances in Visual Computing

View full text Add to dashboard Cite

show abstract

“…Beyond perceptual grouping, several other computer vision tasks have been shown to benefit from a similar inclusion of recurrent processing, including image generation, 67 object recognition, 39,68–70 and superresolution tasks 71 …”

Section: The Role Of Recurrence In Visual Recognitionmentioning

confidence: 99%

Beyond the feedforward sweep: feedback computations in the visual cortex

Kreiman

Serre

2020

Annals of the New York Academy of Sciences

View full text Add to dashboard Cite

Visual perception involves the rapid formation of a coarse image representation at the onset of visual processing, which is iteratively refined by late computational processes. These early versus late time windows approximately map onto feedforward and feedback processes, respectively. State‐of‐the‐art convolutional neural networks, the main engine behind recent machine vision successes, are feedforward architectures. Their successes and limitations provide critical information regarding which visual tasks can be solved by purely feedforward processes and which require feedback mechanisms. We provide an overview of recent work in cognitive neuroscience and machine vision that highlights the possible role of feedback processes for both visual recognition and beyond. We conclude by discussing important open questions for future research.

show abstract

“…The feedback mechanism allows the network to adjust the previous input through feedback output information. In recent years, the feedback mechanism has also been used in many network applications of computer vision tasks [ 15 , 28 ]. For image SR, Haris et al [ 29 ] proposed an iterative up and down projection unit based on back-projection to realize iterative error feedback.…”

Section: Related Workmentioning

confidence: 99%

“…However, the feed-forward method prevents the effective information of the latter layer from being fed back to the previous layer, and the input of the previous layer cannot be adjusted. Hence, recent studies [ 14 , 15 ] have applied the feedback mechanism to the network architecture. In the theory of human recognition, people can always find the correlation between them based on the original data and focus on some of its important features.…”

Section: Introductionmentioning

confidence: 99%

Edge-Enhanced with Feedback Attention Network for Image Super-Resolution

Yin

2021

Sensors

View full text Add to dashboard Cite

Significant progress has been made in single image super-resolution (SISR) based on deep convolutional neural networks (CNNs). The attention mechanism can capture important features well, and the feedback mechanism can realize the fine-tuning of the output to the input. However, they have not been reasonably applied in the existing deep learning-based SISR methods. Additionally, the results of the existing methods still have serious artifacts and edge blurring. To address these issues, we proposed an Edge-enhanced with Feedback Attention Network for image super-resolution (EFANSR), which comprises three parts. The first part is an SR reconstruction network, which adaptively learns the features of different inputs by integrating channel attention and spatial attention blocks to achieve full utilization of the features. We also introduced feedback mechanism to feed high-level information back to the input and fine-tune the input in the dense spatial and channel attention block. The second part is the edge enhancement network, which obtains a sharp edge through adaptive edge enhancement processing on the output of the first SR network. The final part merges the outputs of the first two parts to obtain the final edge-enhanced SR image. Experimental results show that our method achieves performance comparable to the state-of-the-art methods with lower complexity.

show abstract

Feedback Networks

Cited by 148 publications

References 32 publications

Instance-level Object Recognition Using Deep Temporal Coherence

Instance-level Object Recognition Using Deep Temporal Coherence

Beyond the feedforward sweep: feedback computations in the visual cortex

Edge-Enhanced with Feedback Attention Network for Image Super-Resolution

Contact Info

Product

Resources

About