Wheat-Head-Detction
STEP-1
-: dataset-:https://www.kaggle.com/c/global-wheat-detection/data?select=train
-: download-> https://github.com/tensorflow/models
-: filter you image data
-: if there is no BBOX in csv for images then remove those images from dataset
-: csv manipulation
-: now you have (filterd images + csv)
-: create Train.csv and Test.csv
-: create train and test image folder
-: create train.record and test.record
-: I set num_steps: 4310 in "faster_rcnn_resnet50_pets.config" you can change it(initially it is 200000). Try to iterate
more for proper accuracy
-: path to get this file "/models-master/research/object_detection/samples/configs/faster_rcnn_resnet50_pets.config"
-: train your model
-: MAKE SURE, "SET PATH" WHENEVER IT REQUIRES
-: Use this "sys.path.append("../slim")'
-: start training, it will take lots of time
-: Export Graph with "export_inference_graph.py", run this file from your Terminal, Don't forget to put path in the file
STEP-2
-:Use saved .pb file
-:Load it.
-:create rectangle wherever it is detected using OpenCv
-:create CSV for test images (I havn't done this, you just need to dump it into dataframe and then .CSV)