Downloaded 71 times
More Related Content
Similar to Object detection with Tensorflow Api
Object detection with Tensorflow Api
- 1. Object Detection withTensorflow byRayees ahmad
- 2. Agenda ▪ Intro ▪ WhatisObjectDetection ▪ Stateof Object Detection ▪ Tensorflow Object DetectionAPI ▪ PreparingData ▪ Training &Evaluating ▪ Links
- 3.
- 4.
- 5.
- 6. Object detection = ObjectClassification + Object Localization
- 7. One model fortwotasks? Objectdetection-outputistheonenumber(index)of aclass Objectlocalization-outputisthefournumbers- coordinatesof boundingbox. Po bx1 bx2 by1 by2 c1 c2 c3 … cn - is object exists - bounding box coordinates - object’s variables
- 8.
- 9. Approaches ▪ Classicalapproach(Haarfeatures)-first ODrealtimeframework (Viola-Jones) ▪ Deeplearningapproach-nowstateoftheartinOD ▪ OverFeat ▪ R-CNN ▪ FastR-CNN ▪ YOLO ▪ FasterR-CNN ▪ SSDandR-FCN
- 10. Deep learning approach OverFeat-publishedin2013,multi-scalesliding window algorithm using Convolutional Neural Networks(CNNs). al C.NN - RegionswithCNN features.Threestage approach: - Extract possible objects using aregion propos method(themostpopular onebeingSelective Search). - ExtractfeaturesfromeachregionusingaCNN. - ClassifyeachregionwithSVMs.
- 11. Fast R-CNN -Similarto R-CNN, it used Selective Search to generate object proposals, but instead of extractingallof themindependentlyandusingSVM classifiers,it appliedtheCNN onthecompleteimage and then used both Region of Interest (RoI) Pooling onthefeaturemapwithafinalfeedforwardnetwork for classificationandregression. YOLO-You Only Look Once: a simple convolutional neural network approach which has both great results and high speed,allowingfor thefirsttime realtimeobjectdetection. Deep learning approach
- 12. Faster R-CNN -FasterR-CNN added what theycalledaRegionProposalNetwork(RPN), inanattempttogetridof theSelectiveSearch algorithm and makethe model completely trainableend-to-end. SSD andR-FCN Finally, there are two notable papers, Single Shot Detector(SSD)whichtakesonYOLO byusingmultiple sized convolutional feature mapsachieving better results and speed,and Region-based Fully Convolutional Networks (R-FCN) which takes the architecture of Faster R-CNN but with only convolutionalnetworks. Deep learning approach
- 13.
- 14. TF Object DetectionAPI ▪ OpenSourcefrom2017-07-15 ▪ Built on top of TensorFlow ▪ Containstrainabledetection models ▪ Contains frozenweights ▪ Contains JupyterNotebook ▪ Makes easyto construct, train anddeploy object detectionmodels
- 15. Getting started Dependencies: ▪Protobuf2.6 ▪Python-tk ▪Pillow1.0 ▪lxml ▪TfSlim (included) ▪Jupyternotebook ▪Matplotlib ▪Tensorflow(tensorflow- gpu) ▪Cython ▪cocoapi Installation instruction Ifmodelwillbetrainedlocally-better to installtensorflow-gpu. Dependencies for tensorflow-gpu: ▪ NVIDIAGPU withCUDAComputeCapability3.0 (list) ▪ Ubuntu16.04atleast ▪ CUDA® Toolkit9.0 ▪ NVIDIAdriversassociatedwithCUDA Toolkit9.0. ▪ cuDNNv7.0 ▪ libcupti-dev Installation instruction Latest version of CUDA Toolkit -9.1not compatible with tensorflow 1.6,need to install9.0
- 16.
- 17. Dataset ▪ Tensorflow ObjectDetectionAPI usesthe TFRecord fileformat ▪ Thereisavailable third-partyscripts to convert PASCAL VOC andOxford Pet Format ▪ Inother caseexplanation of formatavailable in gitrepo. ▪ Inputdatato create TFRecord -annotated image
- 18. Getting images Grab frominternet ▪ Scrap images from googleor Pixabay orwhatever ▪ Forbatchdownloading - Faktun Bulk Image Downloader ▪ Fordataminingbymultiplying existingimages -ImageMagic Create own images ▪ Recordvideowithneeded object/objects (in640x480) ▪ Process videoandspliton screenshots -ffmpeg Tips ▪ Createimageswith different lights,backgroundandsoon. ▪ If object is able to have differentforms-betterto catch themall. ▪ Tryto make30%-50%of imageswith overlaid object ▪ Tool forimageaugmentation
- 19. Labeling (Annotation) animages Tools ▪ LabelImg ▪ FIAT (FastImageData AnnotationTool) ▪ input:images ▪ output: .xml files with boundingboxescoordinates
- 20. Creating TFRecord ▪ TensorflowobjectdetectionAPIrepocontains folderdataset_toolswith scripts to covertscommonstructuresof datain TFRecord. ▪ Ifoutputdatahasanotherstructure -hereisexplanation howto convertit
- 21.
- 22. Selecting a model TensorflowOD API provides a collection of detection models pre-trained on the COCO dataset,theKitti dataset, andtheOpenImages dataset. - model name correspondsto aconfigfile that wasusedto train this model. - speed -runningtimeinmsper600x600 image - mAP standsformeanaverageprecision, which indicates how well the model performedontheCOCO dataset. - Outputs types(Boxes,andMasks if applicable)
- 23. Training & Evaluating #From the tensorflow/models/research directory python object_detection/eval.py --logtostderr --pipeline_config_path=${PATH_TO_YOUR_PIPELINE_CONFIG} --checkpoint_dir=${PATH_TO_TRAIN_DIR} --eval_dir=${PATH_TO_EVAL_DIR} # From the tensorflow/models/research directory python object_detection/train.py --logtostderr -- pipeline_config_path=/tensorflow/models/object_detection/samples/configs/ssd_mobilenet_v1_p ets.config --train_dir=${PATH_TO_ROOT_TRAIN_FOLDER}
- 24. Applications Of ObjectDetection: • Facial Recognition • Security: • Self Driving Cars
- 25. Facial Recognition: A deeplearning facial recognition system called the “DeepFace” has been developed by a group of researchers in the Facebook, which identifies human faces in a digital image very effectively. Google uses its own facial recognition system in Google Photos, which automatically segregates all the photos based on the person in the image. There are various components involved in Facial Recognition like the eyes, nose, mouth and the eyebrows.
- 26. Self Driving Cars: Self-drivingcars are the Future, there’s no doubt in that. But the working behind it is very tricky as it combines a variety of techniques to perceive their surroundings, including radar, laser light, GPS, odometry, and computer vision. Advanced control systems interpret sensory information to identify appropriate navigation paths, as well as obstacles and once the image sensor detects any sign of a living being in its path, it automatically stops. This happens at a very fast rate and is a big step towards Driverless Cars.
- 27. Security: Object Detectionplays a very important role in Security. Be it face ID of Apple or the retina scan used in all the sci-fi movies. It is also used by the government to access the security feed and match it with their existing database to find any criminals or to detect the robbers’ vehicle. The applications are limitless.
- 28. Links ▪ https://towardsdatascience.com/how-to-train-your-own-object-detector-with- tensorflows-object-detector-api-bec72ecfe1d9 ▪ https://www.kdnuggets.com/2017/10/deep-learning-object-detection- comprehensive-review.html ▪http://www.machinelearninguru.com/deep_learning/tensorflow/basics/tfrecord/tfreco rd.html ▪ https://www.coursera.org/learn/convolutional-neural-networks ▪ https://medium.com/comet-app/review-of-deep-learning-algorithms-for-object- detection-c1f3d437b852 ▪ https://towardsdatascience.com/evolution-of-object-detection-and-localization- algorithms-e241021d8bad ▪ https://medium.freecodecamp.org/how-to-play-quidditch-using-the-tensorflow- object-detection-api-b0742b99065d
- 29.
- 30.