Notes on "Speed/accuracy trade-offs for modern convolutional object detectors"

"Speed/accuracy trade-offs for modern convolutional object detectors" by Huang et al (2017).

The aim of this paper is to evaluate modern convolutional object detection systems where the metrics is speed and accuracy. Separation between "meta-architectures" (Faster R-CNN, R-FCN and SSD) and base feature extractors (VGG, Residual Networks, Inception) are made to evaluate their different combinations.


The paper uses three meta-architectures namely Single Shot Detector (SSD), Faster R-CNN and R-FCN. See the following posts to get familiar with these object detection algorithms. For these meta-architectures six feature extractors are considered and they are VGG-16. Resnet-101, Inception v2, Inception v3, Inception Resnet and MobileNet. Other architecture configurations such as the number of proposals and output stride settings for Resnet and Inception Resnet are also set up in a reasonable way (see the detail in the paper :P). Furthermore the loss function configuration, input size configuration, training and hyperparameter tuning. benchmark on computing and the image size are clearly stated and made for fair comparison.

The results of this investigation in terms of mAP (mean average precision) vs gpu wall clock time for both meta architecture and feature extractor, and memory vs GPI time for only feature extraction are given and summarized below:
  1. Critical points on the optimality frontier: (a) SSD models with Inception v2 and Mobilenet are the most accurate of the fastest models. (b) Faster R-CNN with dense output Inception Resnet models are the most accurate method. (c) Best balance between the speed and accuracy seem to be with R-FCN models using Residual Network.
  2. The effects of adjusting the feature extractor: Strong correlation is found between the classification and detection performance for Faster R-CNN and R-FCN while SDD appears less reliant on the feature extractors classification performance. SSD is also competitive for large objects but is typically poor for small objects.
  3. The effect of adjusting image size: Decrease in image resolution would decrease the accurate but increase the inference time.
  4. The effect of adjusting the number of proposals: The number of proposals can be decreased while achieving similar performance. Increase in box proposals would increase the computational cost but to which degree depends on the feature extractor.
The authors of the paper used ensemble of these models to win the COCO challenge. The main contirbution is the extensive experimental comparison of some of the main aspects that influence the speed and accuracy of modern object detectors - which may help the practitioners of real world applications.


SSD:
Faster R-CNN:
R-FCN:

Q&A:

 Q) YOLO Vs Regional methods. 
SSD is very similar to YOLO  that it uses the same one shot detection.

Comments

Post a Comment

Popular posts from this blog

Notes on "Focal Loss for Dense Object Detection" (RetinaNet)

Image
Focal Loss for Dense Object Detection by Lin et al (2017) The central idea of this paper is a proposal for a new loss function to train one-stage detectors which works effectively for class imbalance problems (typically found in one-stage detectors such as SSD). The authors also introduce RetinaNet based on the proposed loss function which outperforms the state-of-the-art detection algorithms including the two-stage detectors. Focal loss function: pt is defined as p if y = 1 and p-1 otherwise. In this way, CE(pt) is the cross entropy where p is the probability score. As shown below, the cross entropy is denoted blue (gamma = 0). By adding the factor (see the focal loss FL(pt)) the loss function computes smaller value for the well classified examples and puts more focus on hard and misclassified eamples. In this way the easy background examples are dealt with, which enables the training of highly accurate desne object detection algorithms. Just like alpha-cross entropy is the comm
Read more

Introduction to Bayesian methods

Image
Introduction to Bayesian methods: 1. Bayesian way of thinking: A man is running. Why? Some of the possible explanations are (1) this man is in hurry, (2) this man is running, (3) this man just spotted a dragon, and (4) this man always runs. There are three principles in Bayesian way of thinking which we can use to conclude which of these explanations are the most reasonable. These principles are: 1. Use the prior knowledge on the problem - there is no dragon. 2. Choose the answer that explains the data the most - the man is not wearing the runner. 3. Avoid making extra assumptions - crude to think that any man always run. The answer number "1. this man is in hurry" should be chosen following this principles. 2. Review of probability: Checklists are: 1. What is the probability? 2. What are random variables? 3. What is the probability mass function (PMF)? 4. What is the probability density function (PDF)? 5. What does it mean for the random variables X and Y
Read more

Conjugate Priors

Conjugate Priors. 1. Analytical inference: See the Bayes formulae. The problem is that the evidence is difficult to estimate for the analytical inference. One way to get away is the maximum posterior (MAP) and is very easy to compute, but it comes with many problems. 1. Lack of the in-variance to re-parameterization. 2. The MAP cannot be used as the prior. 3. The MAP is an untypical point (the mode). 4. The credible regions may not be computed. 2. Conjugate distributions. Another approach to avoid computing the evidence is called the conjugate distributions. As the likelihood and the evidence is fixed by our model, we can only vary the prior so that it is easier to compute the posterior. I f the posterior distribution p(θ|x) are in the same family as the prior distribution p(θ), the prior and posterior are then called conjugate distributions, and the prior is called a conjugate prior for the likelihood. For example, the Gaussian family is conjugate to itself (or self-
Read more