This solution can be applied to factory defect detection, medical image analysis, biological image analysis, industrial safety image analysis, mask image analysis, etc.
Open the marking software. Prepare a png or jpg image for annotation. It is recommended that the image has the same aspect ratio.
Delete the log folder, which is the record read by tensorboard during training.
Prepare a list of training images.
- image_path = "data/train/images": Training image path.
- txt = "data/train.txt": The output training image list.
To prepare the verification image list.
- image_path = "data / train / images": verification image path.
- txt = "data/train.txt": The output verification image list.
First set the category name to data/labels.txt, the first category name is background, and the second and below are the category names of your samples.
The category name must be exactly the same as the category name when labeling.
Then, run this ipynb to convert the data in data/train and data/val into tensorflow's tfrecord format.
Train the model.
--num_epochs=100000: The number of training periods. num_classes=3: Set the number of categories.
Release the tensorboard that is not in this training.
Turn on tensorboard to display the loss curve and other related information during training.
Infer a picture.
--ckpt_path=model/efficientdet-d0-pcb : Inferred model.
--input_image_size=512 : Image size.
--input_image=data/test/images/capacitor1-4.png : Inferred images.
--num_classes=3 : The number of categories of the model.
Infer the images in the folder.
--input_image=data/train/images/ : Inference folder.
Infer the images in the folder and calculate the correct rate.
Whether it is correct or not is to compare the detected category name with the image file name.
The image file name format is: category name-xxx.png.
Contact Us and How to BuyWelcome to contact us. Please refer to the following link: