Home Artificial Intelligence Education Scalable and Efficient Object Detection

Scalable and Efficient Object Detection

Audio version of the article

Scalable and Efficient Object Detection

As one of the core applications in computer vision, object detection has become increasingly important in scenarios that demand high accuracy, but have limited computational resources, such as robotics and driverless cars. Unfortunately, many current high-accuracy detectors do not fit these constraints. More importantly, real-world applications of object detection are run on a variety of platforms, which often demand different resources. A natural question, then, is how to design accurate and efficient object detectors that can also adapt to a wide range of resource constraints?

In “EfficientDet: Scalable and Efficient Object Detection”, accepted at CVPR 2020, we introduce a new family of scalable and efficient object detectors. Building upon our previous work on scaling neural networks (EfficientNet), and incorporating a novel bi-directional feature network (BiFPN) and new scaling rules, EfficientDet achieves state-of-the-art accuracy while being up to 9x smaller and using significantly less computation compared to prior state-of-the-art detectors. The following figure shows the overall network architecture of our models.


Source link

EfficientDet architecture. EfficientDet uses EfficientNet as the backbone network and a newly proposed BiFPN feature network.

Model Architecture Optimizations
The idea behind EfficientDet arose from our effort to find solutions to improve computational efficiency by conducting a systematic study of prior state-of-the-art detection models. In general, object detectors have three main components: a backbone that extracts features from the given image; a feature network that takes multiple levels of features from the backbone as input and outputs a list of fused features that represent salient characteristics of the image; and the final class/box network that uses the fused features to predict the class and location of each object. By examining the design choices for these components, we identified several key optimizations to improve performance and efficiency:

Previous detectors mainly rely on ResNetsResNeXt, or AmoebaNet as backbone networks, which are all either less powerful or have lower efficiency than EfficientNets. By first implementing an EfficientNet backbone, it is possible to achieve much better efficiency. For example, starting from a RetinaNet baseline that employs ResNet-50 backbone, our ablation study shows that simply replacing ResNet-50 with EfficientNet-B3 can improve accuracy by 3% while reducing computation by 20%.

Another optimization is to improve the efficiency of the feature networks. While most previous detectors simply employ a top-down feature pyramid network (FPN), we find top-down FPN is inherently limited by the one-way information flow. Alternative FPNs, such as PANet, add an additional bottom-up flow at the cost of more computation. Recent efforts to leverage neural architecture search (NAS) discovered the more complex NAS-FPN architecture. However, while this network structure is effective, it is also irregular and highly optimized for a specific task, which makes it difficult to adapt to other tasks.

To address these issues, we propose a new bi-directional feature network, BiFPN, which incorporates the multi-level feature fusion idea from FPN/PANet/NAS-FPN that enables information to flow in both the top-down and bottom-up directions, while using regular and efficient connections.

A comparison between our BiFPN and previous feature networks. Our BiFPN allows features (from the low resolution P3 levels to high-resolution P7 levels) to repeatedly flow in both top-down and bottom-up ways.

To improve the efficiency even more, we propose a new fast normalized fusion technique. Traditional approaches usually treat all features input to the FPN equally, even those with different resolutions. However, we observe that input features at different resolutions often have unequal contributions to the output features. Thus, we add an additional weight for each input feature and allow the network to learn the importance of each. We also replace all regular convolutions with less expensive depthwise separable convolutions. With these optimizations, our BiFPN further improves the accuracy by 4%, while reducing the computation cost by 50%.

A third optimization involves achieving better accuracy and efficiency trade-offs under different resource constraints. Our previous work has shown that jointly scaling the depth, width and resolution of a network can significantly improve efficiency for image recognition. Inspired by this idea, we propose a new compound scaling method for object detectors, which jointly scales up the resolution/depth/width. Each network component, i.e., backbone, feature, and box/class prediction network, will have a single compound scaling factor that controls all scaling dimensions using heuristic-based rules. This approach enables one to easily determine how to scale the model by computing the scaling factor for the given target resource constraints.

Combining the new backbone and BiFPN, we first develop a small-size EfficientDet-D0 baseline, and then apply a compound scaling to obtain EfficientDet-D1 to D7. Each consecutive model has a higher compute cost, covering a wide range of resource constraints from 3 billion FLOPs to 300 billion FLOPS, and provides higher accuracy.

Model Performance
We evaluate EfficientDet on the COCO dataset, a widely used benchmark dataset for object detection. EfficientDet-D7 achieves a mean average precision (mAP) of 52.2, exceeding the prior state-of-the-art model by 1.5 points, while using 4x fewer parameters and 9.4x less computation.

EfficientDet achieves state-of-the-art 52.2 mAP, up 1.5 points from the prior state of the art (not shown since it is at 3045B FLOPs) on COCO test-dev under the same setting. Under the same accuracy constraint, EfficientDet models are 4x-9x smaller and use 13x-42x less computation than previous detectors.

We have also compared the parameter size and CPU/GPU latency between EfficientDet and previous models. Under similar accuracy constraints, EfficientDet models are 2x-4x faster on GPU, and 5x-11x faster on CPU than other detectors.

While the EfficientDet models are mainly designed for object detection, we also examine their performance on other tasks, such as semantic segmentation. To perform segmentation tasks, we slightly modify EfficientDet-D4 by replacing the detection head and loss function with a segmentation head and loss, while keeping the same scaled backbone and BiFPN. We compare this model with prior state-of-the-art segmentation models for Pascal VOC 2012, a widely used dataset for segmentation benchmark.

EfficientDet achieves better quality on Pascal VOC 2012 val than DeepLabV3+ with 9.8x less computation, under the same setting without COCO pre-training.

Open Source
Given their exceptional performance, we expect EfficientDet could serve as a new foundation of future object detection related research, and potentially make high-accuracy object detection models practically useful for many real-world applications. Therefore, we have open sourced all the code and pretrained model checkpoints on GitHub.

This article has been published from the source link without modifications to the text. Only the headline has been changed.

Source link

Must Read

Former Google CEO Eric Schmidt Wants to Create a Government-Funded A.I. University

The U.S. government’s approach of letting Silicon Valley drive the country’s technological boom has left the government itself scrambling for tech talent. Now, a federal commission...

Bitcoin at 12k, Where next ?

After months of anxiety, the Bitcoin price is finally climbing like a rocket. Most of the cryptocurrencies are also performing well. The second most popular...

Is Data the new gold

 Digitalization is yielding vast quantities of data, which offer opportunities for business, human well-being and the environment, if used effectively. New business models...

Understanding Holistic Scene with Panoptic

Real-world computer vision applications, such as self-driving cars and robotics, rely on two core tasks — instance segmentation and semantic segmentation. Instance segmentation identifies the class and...

China Aims to Dominate World Blockchain Industry

The Takeaway:China's blockchain infrastructure BSN is set to provide global access to its services next month. The network’s reliance on U.S. cloud service providers makes...

Integreting Supply Chain with Blockchain

In the course of my 20 years of experience in information technology, I’ve seen many tech trends come and go: big data, the cloud,...

Using Blockchain to support Dairy Industry

Australian dairy farmers are enlisting blockchain technology to regain control of their industry and redress the power balance with processors and retailers.The recent history...

Basics about natural language processing

Learn the basics about natural language processing, a cross-discipline approach to making computers hear, process, understand, and duplicate human speech.It wasn't too long ago...

Is Amazon lenient towards Alexa skills policies

Amazon claims it reviews the software created by third-party developers for its Alexa voice assistant platform, yet US academics were able to create more...

VC’s warns US could lose technology race with China

http://www.podcastone.com/downloadsecurity?url=aHR0cHM6Ly9wZHN0LmZtL2UvY2h0YmwuY29tL3RyYWNrL0UyRzg5NS9hdy5ub3hzb2x1dGlvbnMuY29tL2xhdW5jaHBvZC9hZHN3aXp6LzIwMDIvMDcxNDIwX0xFQURFUlNfQU5EX0xFR0VORFNfcG9kY2FzdF9iNXgxXzk3Y2U0YTJhLm1wMz9hd0NvbGxlY3Rpb25JZD0yMDAyJmF3RXBpc29kZUlkPTU0MDZkZjNhLTgyYWItNGUzZi05ZjVjLTM0MDg5N2NlNGEyYSoqfDE1OTYyNzg5ODMxMzEqKnw=.mp3This week on Leaders and Legends Nick Beim, partner in the Venture Capital Firm Venrock joined Aileen Black to discuss the need to sound...
banner image