sainijagjit/Piling_Machine_Detector

Piling machine detector using yolov3

Piling Machine Detector

Data Scrapping

1. Image Scrapping : File Location customidz/image_scrapping.py Replace variable search_term value with the search string. Replace number of images in search_and_download() function argument at line 142.

It will download images from google and save in images/<search_term> directory. Selenium library is used

2. Video Scrapping: File Location customidz/video_scrapping.py Replace variable search_term value with the search string. Replace number of videos in search_and_download() function argument at line 29. It will download videos from youtube and save in images/<search_term> directory. Youtube API is used

Data

Images dir : darknet/custom_data/images/piling_machine Labels dir : darknet/custom_data/labels/piling_machine

Training images (0.8) : darknet/custom_data/train.txt Test images (0.2) : darknet/custom_data/test.txt

Config File : darknet/custom_data/cfg/yolov3-custom.cfg

Weights dir: darknet/custom_data/backup

Predicted Output : /output

Steps to Test (on Test images)

1. Make the ground truth sheet from labels using ground_truth_sheet.py, it will create a csv file ground_truth_sheet.csv [already created, no need to re-run the program]

2. open detector.py file and do the following changes: • Change cust_dir path variable at line 132 to point to customindz folder. • For getting Confusion Matrix,Accuracy, Recall, Precision, F1_score change change view argument to False in prediction() function at line 133.

   • It will create prediction_sheet.csv file and will calculate the the metrics by calling testing.py file. It will create a file report.csv, where you can see accuracy at different iou thresholds.

   • For viewng test images along with predicted bounding box, change view argument to True in prediction() function at line 133. prediction(cust_dir,view=True)

Detect Your own Image :

1. Open single_file_detector.py and Update Cust_dir variable path to customindz folder at line 102.
2. Give image path in detector() function argument at line 103.

Video Feed:

To see video output, change variable cap at line 28 of existing_video_feed.py file to your video link.