The COCO labels I can detect include stop signs and fire hydrants.
The COCO labels I can detect include animals such as cats, dogs, birds, horses, cows, and sheep.
The COCO labels I can detect include kitchen and dining objects such as wine glasses, cups, forks, knives, and spoons.
I started this repository in 2026-03, and it was last updated in 2026-03: Started 2026-03, last updated 2026-03.
My yolo-coco directory contains the YOLOv3 object detector model files pre-trained (on the COCO dataset).
For image and video-stream detection, I install these dependencies: pip install numpy; pip install opencv-python.
I should not use YOLO when a dataset contains many small objects grouped close together; YOLO uses an SxS grid where each cell predicts only a single object.
For small objects, Faster R-CNN works best for small objects but is the slowest.
SSDs can struggle with smaller objects (but not as much as YOLO); often give a nice tradeoff in terms of speed and accuracy.
I implemented FilesEncryptionDecryption_LinuxKERNEL as a Linux Kernel Module that supports a sys_cpenc system call.
My sys_cpenc system call supports three operations: copy, encrytion, and decryption.
The README identifies two main files containing my code: the user-land file is xhw1.c, and the kernel-land file is sys_cpenc.c.
I started this project in 2026-03, and its last-updated date is 2026-03.
The available project description does not specify the operations’ precise interfaces or behavior, the encryption method, kernel version, build or usage instructions, or test results.
My FilesEncryptionDecryption_LinuxKERNEL repository is on @alokaditya1's GitHub.
I use GitHub as @alokaditya1. My profile describes my interests and location, and my repositories span several languages and project areas.
My GitHub profile lists my interests as machine-learning research; quantum computing and research; cybersecurity; blockchain; macroeconomics; AGI for the environment.
My GitHub profile lists my location and activity note as India; (inactive on GH).
I have been on GitHub since 2023.
My profile lists nine repositories.
Across my repositories, the language counts are Python (4), HTML (2), C (1), Jupyter Notebook (1); one repository has no language specified.
My Atari_2600_DLAgent project supports deep reinforcement learning experiments on Atari 2600 games, using AIGym environments and predetermined hyperparameters, with user-settable program parameters.
I set the default Alpha to 0.0001, the learning rate for the Adam algorithm.
I set the default Gamma to 0.99, the discount rate of subsequent states value.
I set the default Epsilon Minimum to 0.05, the lowest value epsilon can reach.
I set the default Epsilon Maximum to 1.0, the highest and starting value of epsilon.
I set the default Epsilon Delta to 0.000002, the amount by which epsilon anneals each step.
I set the default Replay Size to 150000, the number of experiences stored in memory.
I set the default Update Frequencey to 4, the number of steps between network updates.
I set the default Batch Size to 32, the number of experiences sampled for one update.
I set the default Target Update Frequency to 10000, the number of steps between target network updates.
My Atari_2600_DLAgent repository is described as “Started 2026-03, last updated 2026-03.”
The project “can launch a range of game environments using AIGym, and then train agents how to play based on a pre-determined set of hyper parameters.”
The README lists these requirements: “Python 3.6 or later; AIGym; Tensorflow; Microsoft Build Tools for Visual Studio 2015/2017.”
For GPU performance, the README says: “If you have a NVidia GPU and want to optimise the performance of the program, then the GPU version of tensorflow should be installed.”
After launching the program, it will “ask the user which game they wish to test on.”
The README lists these compatible games: “Alien, Asterix, Asteroids, Atlantis, BattleZone, BeamRider, Berzerk, Bowling, Boxing, Breakout, ChopperCommand, CrazyClimber, Defender, DemonAttack, ElevatorAction, Enduro, FishinDerby, Freeway, FrostBite, Gravitar, Hero, IceHockey, JamesBond, Kangaroo, Kaboom, Krull, KungFuMaster, MontezumaRevenge, MsPacman, NameThisGame, Phoenix, Pitfall, Pong, PrivateEye, Qbert, Riverraid, RoadRunner, Robotank, Seaquest, Solaris, SpaceInvaders, StarGunner, TimePilot, UpNDown, Venture, YarsRevenge, Zaxxon.”
The README identifies the main techniques as “seperate target and value networks, experience replay buffer and batch updating.”
The README describes the CNN as “a scaled down version of a similar structure suggest by deepmind.” It does not specify the layer structure.
Frame preprocessing includes “reducing it from RGB to grey-scale, down-sampling and then stacking 4 consective frame into a 3D tensor.”
The README claims the agent can achieve “super-human results in some games (Pong and Breakout) and achieve some level of learning in most others”; it provides no quantitative scores or test protocol in the supplied material.
The README also describes Stepped Annealing Epsilon (SAE) and Variable Epsilon (VE) as variations for testing different epsilon schedules during training; their instructions and performance write-up are deferred to the project report.
The README points to the “full project report included in the res directory” for further design, rationale, mathematics, and details about the epsilon variations.
conda install git
conda update --all
conda clean -a
pip install git+https://github.com/Kojoley/atari-py.git
pip install gym[atari]
pip install --upgrade tensorflow
To start a default training session, run python agent.py from the project directory, then enter the game name when prompted.
My Human_faces_generation_withDCGAN project uses a DCGAN that generates fake images of human faces.
I trained the DCGAN in my Human_faces_generation_withDCGAN project on CelebA.
For this project, I specified Pytorch and Python 3.6 as the tools to use.
I found that GANs are a bit too sensitive.
I found that trainning them to reach a low loss is a bit challenging.
In my experiments, BCEWithLogits performed better than the mean of squared errors.
I found that setting beta1 to a value between 0.2 and 0.35 seemed to improve my results.
Setting the discriminator learning rate to 4 times greater than the learning rate of the generator helped speed up my training.
I found that setting a small mini batch, such as 16 or 32, also enhanced the results.
I built this Jupyter Notebook project to generate new fake human-face images. The README links to the CelebA project page. It does not give the exact architecture, optimizer, learning rates, number of epochs, measured losses, quantitative evaluation results, or the CelebA data split used.
My FilesEncryptionDecryption_LinuxKERNEL project is a Linux Kernel Module that supports sys_cpenc for copy, encryption, and decryption.
My Object_detection_with_OpenCV_YOLO project performs object detection in images and video streams using deep learning, OpenCV, Python, and YOLOv3 trained on COCO.
My It_helpdesk repository is listed as HTML; last updated 2026-06; no description provided.
My compliance-analyzer repository is listed as Python; last updated 2026-06; no description provided.
I have a Python repository called compliance-analyzer on GitHub. The available project information is limited, so its purpose and implementation details are not yet clear.
I started compliance-analyzer in 2026-06.
The README for my compliance-analyzer repository contains only the heading “# compliance-analyzer”.
I have not documented what the analyzer does, which compliance domain or users it targets, or how it is tested and used.
My interview-notes-agent repository is listed as Python; last updated 2026-05; no description provided.
My Atari_2600_DLAgent repository is listed as Python; 2026-03.
My Human_faces_generation_withDCGAN repository is listed as Jupyter Notebook; 2026-03.
My FilesEncryptionDecryption_LinuxKERNEL repository is listed as C; 2026-03.
My Object_detection_with_OpenCV_YOLO repository is listed as Python; 2026-03.
My pratice repository is listed as HTML; 2024-01; no description provided.
My pratice- repository is listed as last updated 2024-01; language unspecified; no description provided.