Yolo training

card detect
https://www.youtube.com/watch?v=pnntrewH0xg

https://github.com/aleju/imgaug

https://github.com/geaxgx/playing-card-detection card detecion uses opencv, imaug and shapely

https://github.com/Toblerity/Shapely

Soduko
https://www.youtube.com/watch?v=QR66rMS_ZfA

https://github.com/keras-team/keras/blob/master/examples/mnist_cnn.py

http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/

http://norvig.com/sudoku.html

fine tuning
https://groups.google.com/forum/#!topic/darknet/WvBFz4zSSH4 fine tuning

train
http://seangtkelley.me/blog/2017/12/08/training-yolov2-custom-data

https://eavise.gitlab.io/lightnet/notes/02-examples.html

http://www.renom.jp/notebooks/image_processing/yolo/notebook.html

https://blogs.sap.com/2018/07/25/object-detection-with-yolo-for-intelligent-enterprise/

https://timebutt.github.io/static/how-to-train-yolov2-to-detect-custom-objects/

http://guanghan.info/blog/en/my-works/train-yolo/

https://medium.com/@manivannan_data/how-to-train-yolov3-to-detect-custom-objects-ccbcafeb13d2

https://medium.com/@jonathan_hui/real-time-object-detection-with-yolo-yolov2-28b1b93e2088

https://medium.com/@manivannan_data/how-to-train-multiple-objects-in-yolov2-using-your-own-dataset-2b4fee898f17

https://medium.com/@ribomo42/how-to-train-yolo-v2-with-your-own-data-object-and-labels-on-darknet-2b90dbfecb02

http://ashishkhan.com/blog/not-hotdog-app-with-darknet-yolo-face-detection

https://jumabek.wordpress.com/2017/03/04/how-to-train-yolov2-on-costum-dataset/

http://smart-city-sjsu.net/AICityChallenge/papers/NVIDIA_AI_City_Challenge_2017_paper_19.pdf

https://www.youtube.com/watch?v=MISwqExOjEI

33
https://mc.ai/yolo3-a-huge-improvement/

object tracking
object tracking

http://www.cs.toronto.edu/~davidj/projects/towards_real_time_detection_tracking.pdf  Online multi-player  detection and  tracking in  broadcast basketball  videos are significant challenging tasks. In this environments, the target distributions are highly non-linear, and the varying number of objects creates complex interactions with overlap and ambiguities. In this paper, we present a real-time multi-person detection and tracking framework that is able to perform detection and tracking of basketball players on sequences of videos. Our framework is based on YOLOv2, a state-of-the-art real-time object detection system, and SORT, an object tracking framework based on data association and state estimation techniques. For training and testing, we use a given subset of the NCAA Basketball Dataset. As part of the bonus, we trained a two-layer LSTM to do action recognition

https://github.com/nwojke/deep_sort This repository contains code for Simple Online and Realtime Tracking with a Deep Association Metric (Deep SORT). We extend the original SORT algorithm to integrate appearance information based on a deep appearance descriptor. See the arXiv preprint for more information. Uses https://github.com/abewley/sort and https://arxiv.org/abs/1703.07402 paper. Uses https://motchallenge.net/data/2D_MOT_2015/#download multiple object tracking benchmark. https://github.com/rlabbe/filterpy for kalman filtering.

https://github.com/bendidi/Tracking-with-darkflow The purpose of this little project is to add object tracking to yolov2 and achieve real-time multiple object tracking. The current architecture is set to only track one type of objects, but it should be easy to generalise over all objects. Currently support people tracking (as the provided weights for deep_sort were trained on people tracking)

https://github.com/ykamikawa/yolov2-tracking

hacker news
https://news.ycombinator.com/item?id=15956426

https://www.slideshare.net/TaegyunJeon1/pr12-you-only-look-once-yolo-unified-realtime-object-detection slides

ai.stackexchange
https://ai.stackexchange.com/questions/2854/ssd-or-yolo-on-arm

forums
https://ai6forums.nurture.ai/t/questions-on-yolo/251

journals
https://brage.bibsys.no/xmlui/handle/11250/2418432 Recent advancements in machine learning, and in particular deep neural networks, have yielded excellent object detection models. However, these techniques require vast datasets of labeled training images, which are prohibitively labor intensive to produce.

This thesis explores an alternative approach to obtaining labeled training data, namely using 3D models of objects and modern game engines to generate automatically labeled synthetic training data. A simple approach for generation similar to the one used by Peng et al. (2014) is presented requiring minimal user input, making dataset generation virtually free.

https://nurture.ai/top-papers do search with  yolo

gpu
http://gpupowered.org/node/53/ and https://github.com/prabindh/darknet c++ port, euclid labeler (move this to notable forks)

freelancer jobs
https://www.freelancer.co.za/projects/machine-learning/face-recognition-based-yolo-darknet/ Looking for a freelancer with experience in YOLO for Face recognition. your job to create a DLL that will accept image as input, detect the faces in the image and compare between faces its already have in database. if face found in database return JSON of who's face it is. The DLL must also able to learn new faces. each time new face is introduced it should store it and group by same person face.

links
Yolo bounding box

https://groups.google.com/forum/?nomobile=true#!topic/darknet/V4Fza00E-8I From my experience, the yolov3.weights is bounded to 80 classes if it is trained with coco.data. You can check coco.names file. So if you do NOT add any new class es (or labels), it is possible to achieve incremental training based on yolov3.weights. In your case, the possible solution is:

1. Get the index of the original label which you want to enhance on in coco.names file. (e.g. 0 to person) 2. Generate your own additional person training datasets and label txt files, remember to set label index correctly in your label txt file, corresponding to the label index in coco.names (e.g. 0 to person, 67 to cell phone and so forth.) 3. Execute the normal training command (e.g. darknet detector train xxx.data, xxx.cfg) followed by yolov3.weights and -clear flag. 4. After few iterations, the label you care about will get enhanced while other labels' effects will drop dramatically due to the lack of training data.

If you want to add some additional labels, you can try to migrate some original labels to your new labels in coco.names, but do NOT change the total number of classes (80 for coco), otherwise the yolov3.weights loses its value. It's based on my experiences and some experiments.