Mahdi's repositories
UNet-Plus-Plus-Segmention-with-mobilenetv3
Using Unet architecture as decoder and MobileNet V3 as encoder
Age-and-Gender-Recognition
Recognition of face and age and moods using OpenCV
Anomaly-detection
here is a example for Anomaly detection for identification of observations of a dataset
BiSeNet---face---parsing-and-swap
model link will add soon ,we try and train many model for this project . detects are good but fiting the object in background is awful we trying to amendment amendment some part of code to its do its job better
BLHeli-ESC-firmware-test-and-upgrade
BLHeli-ESC-firmware-test and upgrade for Windows test in 12-20-30 and 40 amp ESCs
Custom-CNN-with-EfficientNetV2B0-and-Vision-Transform
Use Breakhis dataset
Discrete-Cosine-Transform-DCT-
Well, now the DCT or Discrete Cosine Transform method In this method, we show the discrete cosine transform of an image as a sum of sinuses with different magnitudes and frequencies. The dct2 function calculates the two-dimensional discrete cosine transform (DCT) of an image, but DCT has the characteristic that for In a typical image, most of the important visual information about the image is concentrated in only a few DCT coefficients. For this reason, DCT is often used in image compression applications, so this is the cliche application. There is a general mathematical formula that you can search on the net. There is one characteristic, and I will send it to you. It has two main dimensions, which are identified by alpha and beta. There is another formula, which, of course, we did not use in the code, to reverse the situation when we use these functions. We use the amount that they give us as a weight, so it can be calculated as a half of our model, but it cannot be used in everything. We have to make a model for each image. The signal that is given from the image is always an 8x8 matrix, whose value is always constant, if this value is not constant. Because our functions have weight, the logic gets messed up and our work is wrong. The exact same formula is used in the code and the summary of the code and no special library is used.
ELUnet---and-his-Encoder
Unet++ with ELU activision as Decoder and NasNet mobile
FER2013-multinet-architecture
Examining different architectures of famous artificial intelligence networks using fer2013 dataset
Genetic-Algorithm
Simple Genetic-Algorithm implementation
GPS-tracker
GPS tracker with gsm
HOG-people-detection
opencv person detection with HOG algorithm(simple)
Monotonic_function-monotonically-increasing
Monotonic function & monotonically increasing problems with LCS and Sequence alignment
Multi-Armed-Bandits
Multi-Stage-Multi-Armed Bandits (MAB) are a class of reinforcement learning problems where an agent tries to maximize its cumulative reward by sequentially selecting actions from multiple options (arms) and observing the rewards associated with those actions.
Naive-Bayes-Algorithm
Naive Bayes Algorithm with NASA datasets
Network-Programming
Socket Programming and Sniffer
RSNA_Brats
Brain Tumor Image segmentation-Brats2019, 2020, 2021
segmentation_models
Segmentation models with pretrained backbones. Keras and TensorFlow Keras.
sentiment-classification
twitter sentiment classification with comments datasets
Siemens-PLC-controller_snap7
In this repository we use snap7 for controlling traffic lights , its use last repository to detect vehicles and with this module we can control lights by 4 condition (busy, normal, free, empty) for more information please read YOLOv7-Traffic-detection-system repository.
Sparse-representations
The parse representations method is generally called a method in which we perform a series of analyzes on the representation we have. We are going to convert it into multi-dimensional arrays, in the form of a matrix. If I want to explain better, we will consider each of these arrays as a signal for our image. Now, what we will do with these signals depends on our project. (The reason for regression is a method of recognizing that now there is no difference in our input data.) Let's create a special order in our signals using the dictionaries that exist for this method. This special order is such that we use a certain coefficient We use very small domains in our matrices. Now, because these changes we gave are only local changes, we will enlarge the domain of the coefficients that we used with the mathematical methods of rolling, but we will not apply these changes to the previous coefficients, we will determine them in their neighborhood. For example, consider two circles that are inside the same circle. Small domains become large circles. There are non-linearities, which are small, but they still create a lot of space in the domain of the main functions, which is what we do with this. It happens that the location of these edges has a specific geometric order, so we use their display in our dictionary. Now, when we have determined the values of these dictionaries in this way, we use this dictionary to place all our signals under the radius, but then again. There are some small approximations that are named as theta t, they are almost the direction of our dictionary. Now if you look at the code, our main functions are exactly like this: show_im show_imgs_results del_patch get_patch fill_patch naive_high_priority_pixel get_boundary_pixels get_dictionary The first two functions have nothing to do with our work The next three functions are related to the division of our signals, one takes, one erases, and one fills The next function identifies the signals with higher priority and finds them The next function comes to find the signal created by demarcation The next function is to create our dictionary using the same data, just pay attention to the value of our dictionary at the beginning of the work. The next function is Inpainting, where we apply the dictionary to all the signals in the image - Now, the application of this case is that, for example, we have an image that has a damaged part, so we cannot use this method on it, we will use deep and machine learning methods to work on a healthy image with the same details of the damaged image. It reconstructs the visual damage image using our model that we trained it
Wearing-face-mask-detection
Face mask detection with mobile net and yoloV4
yolov7-License-plate-detection
Using yolov7 for detect license plate for iranian plates according the last project -> yolo car detction : yolov3 and yolov7
yolov7-license-plate-farsi-OCR
Using yolov7 for farsi OCR license plate for iranian plates according the last project -> yolo plate detection : yolov3 and yolov7
YOLOv7-Traffic-detection-system
Traffic detection system using YOLOv3 and YOLOv7 for manage road intersection, fully compatibility with serial and IP Relay for traffic light
YoloV9-Smoke-detection
Smoke detection in two classes (white, black) with YOLO V9