Questions tagged [object-detection]
Object detection deals with recognizing the presence of objects of a certain semantic class (e.g., “humans”, “buildings”, “cars”, etc.) in digital image and video data.
208
questions
0
votes
0
answers
8
views
Object detection: better to train with imbalanced dataset or remove images to balance out [duplicate]
I am training an object detection model using the YOLOv5 architecture. I have the following classes and counts.
...
- 239
1
vote
0
answers
10
views
Probability of an image containing a specific object, by combining the results of multiple dependent tests
I am trying to assign to images the probability of them containing a metal building. The images can contain either metal buildings, non-metal buildings or no building.
Let $B$ be the event that an ...
- 11
0
votes
0
answers
14
views
Confidence intervals for Object Detection metrics
I would like to come back on this "When do we require to calculate the confidence Interval?" since recently a reviewer asked me to provide confidence intervals for metrics regarding my work ...
- 101
1
vote
0
answers
9
views
Any image datasets with Inter-annotator agreement (IAA) values recorded? [closed]
Is anyone familiar with publicly accessible image datasets that include Inter-annotator agreement (IAA) scores? Ideally for object detection or classification tasks.
Thank you!
- 11
0
votes
0
answers
13
views
Drop-out in AlexNet [duplicate]
In the section of AlexNet where Drop-out is explained, it's mentioned that Drop-out is not performed during testing, and instead, the output of the neurons is multiplied by 1/2. Does this mean that ...
1
vote
0
answers
22
views
Question about AlexNet
I'm curious about the reason why the third convolutional layer is connected to the feature maps of both previous stages.
Additionally, I wonder why the third and fourth convolutional layers don't ...
1
vote
0
answers
19
views
Question about Color Histogram [closed]
In the description of the Color histogram, it is stated that 25 bin is set for each color channel, the dimension is n = 75 (RGB 3 channel * 25 bin), and then L1 Normalization is performed. Does this ...
0
votes
1
answer
28
views
Distance measurement methods used in Hierarchical Clustering
In Hierarchical Clustering, what are the distance measurement methods used? Are different measurement methods used depending on the purpose?
If performing Hierarchical Clustering for Region Proposal, ...
0
votes
0
answers
36
views
How to calculate Mean Average Precision at a fixed IoU threshold?
Papers such as COCO and the VisDrone often list MAP, MAP50, MAP75 in their evaluation criteria. To my understanding, MAP is the averaged AP for all classes across an IoU range (0.5 - 0.95 for example),...
0
votes
0
answers
17
views
Data augmentation to equilibrate class occurences on image set with multiple different objects in each image
I am using this (reduced) dataset to run tests for detection of defects in solid wood using YOLOV5. Initial tests are promising. As can be seen in Fig. 1, my class occurences are quite imbalanced - ...
0
votes
0
answers
9
views
Should I scale the bounding boxes after padding?
Say if I have a input feature with length 50 and the ground truth bounding box is [10, 20]. If I pad the input to fixed-length 100 and mask the padding in the model, then in the output, should I make ...
- 281
0
votes
0
answers
17
views
How to check statistical significance when comparing four different datasets with same sample size and classes tested on same object detection model
I have results of testing four different image datasets that are identical in all aspects sample size number of classes, number of images for each class etc. except they were captured using different ...
- 1
0
votes
0
answers
19
views
Average precision in in calssification vs in object detection
I think I understand what average precision is: the area under the precision-recall curve.The curve is constructed by calculating the precision and recall metrics at each threshold. There are a few ...
0
votes
0
answers
15
views
How can I evaluate the performance of a object detector at a fixed confidence threshold?
I have an object detector and now I have to decide which confidence threshold to use for each class. How can I determine what is the best confidence threshold for each class? Once decided, how can I ...
- 1
0
votes
0
answers
22
views
What is anchor-based and anchor free two detection paradigm in object detection?
I was reading this paper and saw this passage:
The architecture of these detectors has evolved from the initial two-stage [9, 26, 3] to one stage [19, 31, 1, 10, 22, 13, 36, 14, 7, 33, 11], and two ...
- 1,045
0
votes
0
answers
23
views
Which is more accurate, semantic segmentation or object detection?
In my project, which is product surface defect detection, such as to detect scratches on the washing machines, the target is to increase the precison and recall of detection.I think object detection ...
- 1
1
vote
0
answers
19
views
Object detection and geolocating the bounding boxes
I have acquired videos from a dashcam installed on a vehicle that went around different portions of an airport. My objective is to detect cracks in the pavements. After identifying them, I then need ...
0
votes
0
answers
114
views
Is my approach right for simple single-class object detection?
I want to train a CNN to detect the position of an object in the image. Given an input image, I know that the image can contain either 0 or 1 instances (i.e. one class, no more than one item per image)...
- 235
0
votes
0
answers
53
views
Huber loss problem for object localization
Object localization refers to finding a bounding box for an object in a frame for the purpose of object detection.
I have read here on page 2 that Huber loss has a problem in object localization as it ...
- 249
0
votes
0
answers
44
views
Downsampling feature map to detect small objects
Given $Model$ detection neural network that could be YOLO, EfficientNet, etc. It's recommended that if the detection model $Model$ has a problem detecting objects at a far distance to:
increase the ...
- 249
0
votes
0
answers
10
views
Objects Color Matching against a Reference Standard (ColorCODEX)? [duplicate]
Im trying to build and train a Machine Learning model that autonomously perform color matching between the target gemstone and the Reference Standard color chart. A digital photo image of the target ...
- 173
2
votes
1
answer
445
views
Average Precision (AP) for object detection, huge confusion
I've been reading about how object detection models are evaluated. It seems that the metric most often used is AP. But I have stumbled upon 2 different approaches that I think mean completely ...
- 37
0
votes
0
answers
70
views
Custom Object Detection TensorFlow Lite model
I am trying to develop an Object Detection Model which detects the trained object in the image attached below. I have 20 different patterns of such image, which all all are unique in all four ...
0
votes
0
answers
127
views
I want to detect human hands with gloves on using yolov5. What approach should I use?
I would like to use yolov5 to detect a human hand with gloves on.
However, few such images exist in COCO or OpenImagesDataset.
The same is true for human hand annotation data such as oxford-hand-...
- 21
0
votes
0
answers
105
views
How to reduce the number of false positives in object detection in fusion pipeline?
I am training a sequential fusion network using first in image and then in LiDAR point cloud. Specifically, I tend to using the result of image detection result, to improve the performance for far ...
- 1
2
votes
1
answer
240
views
For "fine-tuning", does the "domain adaptation" approach make sense?
I understand "domain adaptation" to be a type of "transfer-learning" technique.
Domain Adaptation:
By applying knowledge obtained from a domain with sufficient teacher labels (...
- 21
1
vote
0
answers
110
views
YOLOv1 and YOLOv2 -- Pr(Object) and "responsible" boxes
In the YOLOv1 paper, Pr(Object) denotes the probability that there is an object in a certain grid cell.
When it comes to the ground truth values, which values can Pr(Object) take? Only 0 and 1?
What ...
- 35
0
votes
0
answers
10
views
Help training: few-shot segmentation with dense objects
I'm looking for help in how to be able to train supervised segmentation models with a relatively few number of annotations on dense regions. As an example, consider this image of bacteria. While I ...
- 101
1
vote
2
answers
944
views
How to count number of false positives and false negatives in object detection
I've got a rather simple question for you, however to which I can't find a proper answer. Let's have a simple setting which is visualized in the image below. We've got one object of class A in the ...
- 21
1
vote
0
answers
185
views
Guidelines for using NMS before calculating mAP for object detectors
I am having a hard time understanding how to use Non-Max Suppression (NMS) when trying to evaluate an object detection model, especially when paired with trying to calculate metrics like the mean ...
- 153
1
vote
0
answers
749
views
The alpha parameter of focal loss
I want to use a weighted focal loss for my imbalanced object detection problem.
\begin{equation}
L = - \alpha(1-\hat{p})^\gamma log(\hat{p}) , \ \hat{p} = \begin{cases}
p, & \text{if}\ ...
- 11
4
votes
1
answer
702
views
Training with extremely imbalanced Dataset
I have a object detection problem which has extremely imbalanced dataset. Lets say there is only one class to detect, say banana or not banana. This detection network will be used in a real case where ...
- 41
3
votes
0
answers
372
views
I am trying to do object detection using Yolov5 and modify the model. How can I do that?
I want to modify Yolov5 and see how it performs. Here's the yaml file https://github.com/ultralytics/yolov5/blob/master/models/yolov5s.yaml
How do the make the changes in the yaml file if I want to ...
- 31
0
votes
1
answer
212
views
Can bounding boxes be used in UNets?
I recently read a paper where the researchers used a UNet algorithm to localize/detect cyclones using a bounding box. However, my interpretation of a UNet is that it performs semantic segmentation and ...
- 1
1
vote
0
answers
137
views
Yolov3 targets build, Do all 3 scales need to have an anchor box designed for each object in the image or just the scale with better IoU? [closed]
I'm implementing Yolo v3 in Pytorch,
Imagine we have an image with a large object in it, whem building targets, I'll have to assign the anchor box with highest IoU with this object for each scale, (So ...
- 11
1
vote
0
answers
15
views
"Biasing" a generalized trained object detector towards specific examples during inference
I have trained an object detection deep learning model on many different types of cars (shape, color, car model variations etc.). I'm just using a single class "Car" for all the different ...
- 11
1
vote
0
answers
490
views
YOLO v2 loss function
I'm trying to understand (and implement) the YOLOv2 loss function, which is not given explicitly in the original paper.
There are several posts on this topic, but quite a few seem to confuse the ...
- 11
1
vote
2
answers
427
views
Why do we sort confidence scores when calculating MAP in Object Detection?
This question is specific to object detection's metric Mean Average Precision.
I sieved through multiple articles like the one by Jonathon and here by Kaggler Tito.
One part I cannot grasp immediately ...
- 745
0
votes
1
answer
671
views
What should I do with Images that has no objects?
I have a dataset that contains images that has cancerous nodules. I want to use object detection models to detect these nodules from the image by using an object detection model. Now my dataset has ...
1
vote
1
answer
747
views
Is Intersection over Union in object detection differentiable?
Is IoU in object detection differentiable or can be back propagated? Is it used in the training process or just in the inference?
1
vote
0
answers
74
views
Annotations and Bounding Box Definition in Object Detection
I'm having confusion about the concepts of object detection.
If the dataset has been annotated to have ground truth in (x, y) of the four vertices, then why are the papers (using the same dataset) are ...
1
vote
0
answers
26
views
How are the region proposals processed in Fast R-CNN during training and testing?
How are the region proposals processed in Fast R-CNN during training and testing?
From my understanding, during training we sample some small number (e.g. 64) of region proposals out of the ~2000 ...
- 31
1
vote
0
answers
78
views
Conditional class probabilities in YOLOv1
In YOLO v1, authors defines conditional class probabilities (P(Class_i/Object))
I don't understand that authors can define the head output to conditional class probabilities. When the model finishes ...
- 11
1
vote
1
answer
2k
views
When is Non-Max Suppression used in Object Detection
Is non-max suppression for bounding boxes obtained from a Region Proposal Network performed during training? From what I gather, NMS is not differentiable-- in which case, it can't be performed during ...
- 23
1
vote
0
answers
77
views
how to check if unknown object partly covers the object i want to detect
I want to detect an object from an image,I use yolo object detection. I detect car from the image. I consider the case when another unknown object partly covers it, how can my program know that there ...
0
votes
1
answer
944
views
Object detector evaluation (multi-class)
I have a (long) question regarding the evaluation of object detectors. I have two separate networks, the first one detects relevant objects (-> bounding boxes) and the second one classifies the ...
1
vote
0
answers
82
views
Fine-tuning a pre-trained Tensorflow model
I have two questions regarding fine-tuning a pre-trained Tensorflow model on a mix of high-quality and low-quality images.In other words, '
Does the large contrast in image quality confuses the model ...
- 11
0
votes
0
answers
471
views
Template matching for detecting multiple objects in image (+500 different categories)
I want to detect items from the inventory of a game. Detection is in real time, so I need decent performance (> 5 FPS). There are over 500 different items, but the item icons will not change so a ...
- 1
1
vote
0
answers
50
views
Summary of all the results of all participants after 2015 in ImageNet (ILSVRC) challenge. Does such resource exist?
I have been reading about the ImageNet Large Scale Visual Recognition Challenge (ILSVRC). And while I can find the paper by Russakovsky et al. from 2014(updated in 2015), which contains all the ...
- 63
1
vote
1
answer
361
views
Performing object detection without localization
I'm looking to build a computer vision model that can detect the presence or absence of several classes of object within an image, but the localization of these objects is not required. The end ...
- 13