Product detection

This code repo tackle the product detect problem. Given an image, the trained model needs to predict the category it belongs to.

Hyperparameter adjustment

Most of hyperparameters are configured inside main.py, the main script to train/evaluate model.

• Number of epochs -> FLAGS.epoch
• Batch size -> FLAGS.batch_size
• Working dir -> FLAGS.working_dir
• Image input size -> FLAGS.im_size
• Training data path -> FLAGS.training_data
• Test data path -> FLAGS.test_data
• Learning rate -> learning_rate, default to 1e-4
• Starting learning decay iteration -> lr_start_decay, default to the 30-th epoch
• Learning rate decay frequency -> lr_decay_every, default to every 10 epochs
• Data augmentation mode -> FLAGS.augmentation, default to True
• TF records -> FLAGS.use_tfrecord

Run training and monitoring

Execute:

python main.py

To monitor the training process:

tensorboard --logdir=model/summaries --port 5000

and nagigate to localhost:5000 on the web browser.