People who code: we want your input. Take the Survey

# Tag Info

5

This is just an idea Given a set of pixels, the task is to decide: Which pixel is the center of an object? What is the size of the bounding boxes with the center is the pixel in part 1? Formula, consider this is a 2D image, call $(x,y)$ is the horizontal and vertical coordinate and $(w_i,h_i)$ is the size of bouding box of object $i$: \$\text{For }m \in[x,x+...

3

Yes, it is not specified because the region proposal algorithm did not change from R-CNN (the previous version from Fast R-CNN, however, in the next verion, Faster R-CNN, this algorithm is replaced by a CNN). The region proposal algorithm you are looking for is called selective search. You can find in the R-CNN paper that the algorithm is described in "...

2

This is really a semantic segmentation problem if OP wants to pinpoint the weeds. If OP wants to have such a segmentation he will need to hand segment every.single picture.

2

If each photo is intended to show a photo of weed or crops you should give one label. If your task is different where you also try to localize weed or crops in the image, then you need to label accordingly. My understanding is you are trying to do the first case, therefore, there should be one label for each image.

2

The output of YOLO is (x,y,w,h,confidence,class). The confidence value presents whether the rectangle holds an object, the rectangle is non-classed when confidence is low. The class value will be used, only when confidence is high.

2

There are different questions and even different lines of thought here. Let's go through them On resizing Why do we need to resize? To fit the network input which is fixed when nets are no Fully Convolutional Networks (FCN) What if my net is FCN? Still makes sense to resize to bound the dimension of the input features you want to detect (a person on a small ...

2

As per your requirements, I would suggest that you start with any simple CNN network. CNNs take advantage of the hierarchical pattern in data and assemble more complex patterns using smaller and simpler patterns. Therefore, on the scale of connectedness and complexity, CNNs are on the lower extreme. Here is a Keras example: model = models.Sequential() ...

2

To verify the accuracy of the classification stage, you will need labeled images with a single car. To train and verify accuracy of the detection stage and full system, you can: in the datasets with images with multiple cars, manually, mark the image rectangles that contains one car. from previous, split the image in one or more ones, each one containing a ...

2

Unfortunately, the answer here is that "it depends". People have taken different approaches to this problem and I'll describe a few here. None of which however is the "right" answer. Labeling When generating benchmark datasets, we actually do have this problem. To be honest, most of the time the labeling is done to the best ability of the ...

2

In my opinion, it is not because ensemble methods are not good, just the state-of-the-art and Kaggle competitions are two different fields. Kaggle competitions can be understood as an industry project where the target (accuracy, distance value, etc) is the most important, and they can select some computationally expensive way such as ensemble methods to ...

2

I would look at table 1 of the original paper. While you're reading the alogorithm, try to really focus on Step 2 when you get to it. In summary, each feature is used to train it's own classifier. So in your example, the calculated features X1, X2, ... Xn you describe coorespond to apply some set of feature transforms f_1, f_2, ... f_n to a single image. ...

1

As you ask, "in general...", I will answer generally, however this changes a lot from model to model and the way they handle close objects. In general, yes, they would do a poor job detecting very close objects, switch to segmentation models for that (for class or better, instance segmentation). In general, objects detectors learn to tell an object ...

1

I might be able to help with the theory, but the coding... it is a non standard API such as Tensorflow or Pytorch (it might be custom code for what I can tell). The key element here is that the bouding boxes are removed only if they hold a prediciton for the same class that the box that is overlapping with (but with less confidence, that is why it gets ...

1

You are correct about your first assumption, but not about your second assumption. More layers does not always mean better pattern detection. The analogy that in the deeper layers the network learns more complex features is an oversimplified one. It is true to some extent, although it is not enough to explain very complex architectures like GoogLeNet. ...

1

Object detection models usually generate multiple detections per object. Duplicates are removed in a post-processing step called Non-Maximum Suppression (NMS). The Pytorch code that performs this post-processing is called here in the RegionProposalNetwork class. The filtering loop you've mentioned performs the NMS and applies the score_thresh threshold (...

1

Good question! Using Yolo to recognise characters would be a good experiment to try. It may be because of the density of characters on a page -- systems like Yolo are very good at detecting a small number e.g. 2,3 or 10, objects, but don't work so well when the number of objects is the hundreds as you might have with OCR. A better approach might be to try ...

1

Yes, the functionality should is there. But, don't you think you are overdoing the scales. You have at least 18 scales mentioned here. Too much of anything is bad. There is a reason it likes things divisible by 32 because at that increase in size something more meaningful will show up in the image. Spamming sizes like this won't help you at all, it would ...

1

The Standard Image Captioning Pipeline is to train the model in a single batch(or mini-batch) i.e. get the features from the CNN Image encoder and then feed that into an RNN decoder (features + Real Captions) to produce output captions for the Image. The training loop in PyTorch would look something like this: # zero the parameter gradients decoder.zero_grad(...

1

There are two related problems for images Semantic segmentation, where you need to assign each pixel on the image some class. I.e. you have a satellite image and want to segmentate roads/forests/fields and so on Objects detection, where you need to detect different types of objects and draw a bounding box for each. I.e. there is a popular dataset MSCOCO for ...

1

If you just need to draw a rectangle around each key, this is an object detection or template matching problem, so you can use any of the available models for object detection (e.g. YOLO) or any technique for multi-template template matching (e.g. you can use sequential RANSAC or t-linkage). In the first case, you will need a labeled dataset, while, in the ...

1

To have a more powerful representational output. You can derive a bounding box from heatmap but not vice-versa. Also, in case of dense object detection it is hard to create bounding boxes for each object (people standing in front of each other). That being the case, it is better to run a segmentation loss for these networks. It also leads to less confusion ...

1

You can reduce your photo size and scale the corresponding boxes to the new dimensions (416x416). Or if you want to go with your technique, you can slice the image and then, check if the bounding box lies in the slice, then, reorient it according to the slice you took. Take a look at albumentations library for this.

1

You can find an explanation here (github of the googleapi): My current understanding of a color's score is a combination of two things: What is the focus of the image? What is the color of that focus? For example, given the following image: The focus is clearly the cat, and therefore the color annotation for this image with the highest score (0.15) will ...

1

The Problem We can see from the question that existing information on detection and classification in the small automotive vehicle domain has been located (in the form of two independent sets of vectors usable for machine training), and there is no already existing mapping or other correspondence between the elements of one set and the elements of the other. ...

1

That depends! You can try it. But if you change the sizes, you have to ensure that you do not mismatch the shapes. As far as size is concerned it won't affect the accuracy much unless you are significantly changing it. Since the default is 1024x1024, and you are making 1100x1100, there wont be any issues. Remember, there is a tradeoff in terms of speed and ...

1

There is a paper face pose estimation It uses a very straight forward technique, and very obvious augmentaions to achieve nice results. You could use exactly the same if you have a tagged dataset for cars rather than for faces. I was able to reproduce the results myself a while back.

1

I think the problem can be phrased (more generally) as a Pose Estimation Problem. That term might help in obtaining better search results when searching for relevant papers. One paper that I found on the given topic was this one. Even if it is maybe (for whatever reason) not what you are looking for precisely, it might still contain valuable references to ...

1

The neural network will learn what we teach it, for example with that image only, after finish training, your model will hard to recognize humans with dark skin, glasses, big eyes, etc, the features that two annotated targets don't have. If your data is big enough, and contain all the feature of humans face, the result should be good. If not, I recommend a ...

1

Google recommandation seems to answer this: The training data should be as close as possible to the data on which predictions are to be made. For example, if your use case involves blurry and low-resolution images (such as from a security camera), your training data should be composed of blurry, low-resolution images. In general, you should also consider ...

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