jangdonghae/embedded-system-project---binary-quantization-reactnet-base-

Embedded-System-project---binary-Quantization-Reactnet-base-

To run Reactnet for cifar10

1. go to mobilenet/1_step1 directory
2. run run.sh, or execute python train.py --batch_size=128 --learning_rate=5e-3 --epochs=60 --weight_decay=1e-5
3. go to mobilenet/2_step2 directory
4. run run.sh, or execute python train.py --batch_size=128 --learning_rate=5e-3 --epochs=35 --weight_decay=0, after making /models directory and put the model trained on step1

To run Reactnet with cutmix(data augmentation technique)

1. go to mobilenet/1_step1 directory
2. run run_cutmix.sh, or execute python train_cutmix_binary.py --batch_size=128 --learning_rate=5e-3 --epochs=60 --weight_decay=1e-5 --beta 1 --cutmix_prob 1
3. go to mobilenet/2_step2 directory
4. run run_cutmix.sh, or execute python train_cutmix_binary.pyy --batch_size=128 --learning_rate=5e-3 --epochs=35 --weight_decay=0 --beta 1 --cutmix_prob 1, after making /models directory and put the model trained on step1

To run Reactnet with mixup(data augmentation technique)

1. go to mobilenet/1_step1 directory
2. run run_mixup.sh, or execute python train_cutmix_binary.py --batch_size=128 --learning_rate=5e-3 --epochs=60 --weight_decay=1e-5 --beta 1 --cutmix_prob 1
3. go to mobilenet/2_step2 directory
4. run run_mixup.sh, or execute python train_cutmix_binary.py --batch_size=128 --learning_rate=5e-3 --epochs=35 --weight_decay=0 --beta 1 --cutmix_prob 1, after making /models directory and put the model trained on step1

embedded-system-project---binary-quantization-reactnet-base-'s Issues

How to use the trained model

Dear author, thank you for sharing your code generously, but I still have a problem I can't solve.
After run the ReActNet-cifar10\mobilenet\1_step1\train_cutmix_binary.py, I have got "model_best.pth.tar", and I want to use the trained model to predict a new picture, but I don't know how to write the "predict" code.
Could you please share your "predict" code too?
All the best wishes for you.