22

To answer your questions: Yes your approach is right Of A, B and C the right answer is B. The explanation is the following: In order to calculate Mean Average Precision (mAP) in the context of Object Detection you must compute the Average Precision (AP) for each class, and then compute the mean across all classes. The key here is to compute the AP for each ...


18

Other answers suggest to put an additional channel, I disagree. I think it's a very computationally intensive, time consuming process. Moreover, it forces non-pixel data to be processed by Conv filters, which doesn't make much sense IMHO. I suggest you to establish a multi-input model. It would be composed by three parts: A Convolutional part, to process ...


15

There is a nice and detailed explanation with an easy to use code on my Github. Certainly it will help you guys.


11

Although many solutions in production systems still use a sliding window as described below in this answer, the field of computer vision is moving quickly. Recent advances in this field include R-CNN and YOLO. Detecting object matches in an image, when you already have an object classifier trained, is usually a matter of brute-force scanning through image ...


10

I'd want to add @Neil_Slater's answer by sharing my application. In my application, I want to train a model that can automatically load a chess position from a chess book like this: Before I did anything, I made sure I had a model that can accurately detect a chess piece. It was not a hard-problem because it was like training the MINST digits. I collected ...


8

Hashing is the way to go if you want fast -- constant time -- retrieval of nearest neighbors. Here's a recent example using neural networks to learn a binary hash: Deep Learning of Binary Hash Codes for Fast Image Retrieval (code) (slides) You want to avoid doing all-pairs computations like correlations.


8

This is a huge topic, so I will just give you a high-level overview and some pointers to more info. Yes, there are definitely ways to work with video that are different from working with individual still images. At the simplest level, it could be running an object detector such as HoG (or a sliding-window convnet) on each frame, and then some means of ...


7

Yes, there are models that do this. This link points to one of the first papers I believe. The main idea is called weakly supervised object detection. The paper essentially makes three modifications. They treat the typical hidden fully connected layer as a convolutional layer. This works because convolutional layers can be thought of as convolving the same ...


6

A very clear and in-depth explanation is provided by the slow R-CNN paper by Author(Girshick et. al) on page 12: C. Bounding-box regression and I simply paste here for quick reading: Moreover, the author took inspiration from an earlier paper and talked about the difference in the two techniques is below: After which in Fast-RCNN paper which you ...


6

Edit: after the edit in the question, 1) does not relate so much anymore, but 2) still does. It depends a bit on the form of the location data. If you have a segmentation mask (i.e. another image with two colors denoting for each pixel if it belongs to the object or not), then going with another channel as n1k31t4 suggested might be a good idea. If you ...


5

You are using the training set that opencv is giving you which it doesn't correspond to the kind of images you are using. The data you are using comes from getDefaultPeopleDetector and the kind of images that the default detector uses are pictures of many people, not a female model from a fashion ecommerce. If you want to distinguish between models and ...


5

The simplest thing to try out is to put the information in an extra channel of the image. So if you have RGB channels, you could add a four channel which would simply be the location information you have, repeated for every pixel. This creates a lot of redundancy, of course, but means you can take any standard image classifier and it will still work.


4

The paper cited does not mention linear regression at all. What it does is using a neural network to predict continuous variables, and refers to that as regression. The regression that is defined (which is not linear at all), is just a CNN with convolutional layers, and fully connected layers, but in the last fully connected layer, it does not apply sigmoid ...


3

Each grid predictor in YOLO should only have a high score that an object is within it, if it detects the centre of the bounding rectangle is inside itself. So a grid point that contains only the wing mirror of a car should decide it has a low probability of containing the centre of the car. The predicted bounding rectangles are not constrained in the same ...


3

Another approach is "Training object class detectors using only human verification" We propose a new scheme for training object detectors which only requires annotators to verify bounding-boxes produced automatically by the learning algorithm. Our scheme iterates between re-training the detector, re-localizing objects in the training images, and ...


3

The question itself is not quite clear, since you don't state that you have a model that can detect one car per run for an image or you are just asking what tools, algorithms or frameworks to use to detect cars (or other objects) in an image. Answering second variant, you should be using developed algorithms for object detection, which are either Haar ...


3

My doubt is whether the resize would now change the position of my object according to annotation? Yes, it will. should i annotate on the resized images(resize them myslef before training?) No, you should annotate at the original size. You solve this by applying the corresponding transformations on your bounding boxas well. So if you resize your ...


2

The framework you cope with is semi supervised. You have mostly unlabelled data and you can have some labeled data by manual labelling. Active learning is one method to cope with the situation, by focusing your labelling efforts in the most beneficial areas. You can read a survey on these techniques at Settles, Burr (2010), "Active Learning Literature ...


2

One major issue with this approach is that if you represent your image as a vector of pixel intensities, you have zero translation-invariance. Two pictures that are exactly the same but shifted over 10 pixels will likely look like dissimilar vectors. That is why people try to use other methods that handle translation, like convolutional neural networks or ...


2

Technically this is possible but most approaches combine localization with some sort of classification so that the network can get better at scoring regions, and also removes the problem of having to use the [0, 0, 0, 0] vector to encode a null region since if the classifier associated with the region gives a low probability for that object you can ignore ...


2

You can take a look at the YoloNet which detects Objects based on Pascal VOC 2012 dataset. Here is the link to it : https://github.com/pjreddie/darknet/wiki/YOLO:-Real-Time-Object-Detection You should be able to detect vehicles with this pre-trained model.


2

Object detection models (such as SSD, Faster-RCNN, YOLO, R-FCN) are trained to detect specific classes. If you wish to detect a single class, you could train a custom model on this class.


2

In general, you want to show your model all types of "resistors" it would see in the real world. Is it feasible for you to manually go through your dataset and label each of these images as resistor A or resistor B? Do you care if your model can make this distinction? If the answer to either of these is no, I would just leave them all as one label and ...


2

If you only need to predict whether an object is a pepper or not why bother predicting whether it is a red pepper or a green pepper, you are making your problem more complicated than it needs to be. In my opinion you should try to make the problem you want to solve as simple as possible so I would advise using the shortest label possible as long as it solves ...


2

What way I could follow to achieve this? Create a dataset with image-label pairs. Create a classification model. I don't think you will find any algorithm "out-of-the-box" for you to use. You might have a look at some paper to see the kind of approaches they have. I would recommend you look at these: https://arxiv.org/ftp/arxiv/papers/1802/1802.04903.pdf ...


2

In the specific case of knowing the location of the object in the image, one technique would be to crop and pad each training example so that the object is in the exact center. This way the extra information is passed to the neural network implicitly. This is how most face identification neural networks work. If the "location" of the object is more ...


2

I will definitely try one more approach in this case, which is explained below. Will use simple CNN architecture, followed by fully connected layers. Say, now I have fully connected layer(FL) of size 100. Using this FL apply another liner regression model(followed by activation layer...). Structure of linear regression would be: y = w1i.FL(Ni)+ w2i.f1+ w3i....


1

To build a model to detect the old categories and your new ones, you need to re-train the model with your own dataset and the dataset used to pre-train the model. Fortunately, this dataset is available to the public for downloading. The dataset used to train the model is written in the name of model you choose. It's probably the COCO dataset. Since the ...


1

As well as the algorithm used for region proposals in R-CNN and in Fast R-CNN, RPN task in Faster R-CNN is to propose regions that may or may not be an object, so its cls head with 2k scores corresponds to the classification of every anchor as an object or as a non object (this way all of them may be non objects): But it could also be a k scores logistic ...


1

There are multiple solutions: Do linear scalarizing. E.g. If you have $d_0$ and $d_1$ then you could do: $$D = w_1 d_1 + w_2 d_2$$ You could use the Pareto optimal front. The Pareto optimal front is the set of all distance pairs where one of the distances cannot be dominated without deteriorating the other one. This solution is probably not that well ...


Only top voted, non community-wiki answers of a minimum length are eligible