MobileNetV3 tf.keras implementation with pre-trained weights using the TPU Estimator API.
I have tested this implementation using tensorflow 1.13
Top-1 Acc. | Top-5 Acc. | Path to weights: | |
---|---|---|---|
V3Large | 75.0% | 92.0% | Google drive |
V3Small | 67.3% | 87.4% | Google drive |
Please see main.py
for detailed information on available flags.
export TPU_NAME=<your TPU name>
export MODEL_NAME=MobileNetV3Small
export STORAGE_BUCKET=<your imagenet bucket location>
export DATA_DIR=${STORAGE_BUCKET}
export OUTPUT_DIR=${STORAGE_BUCKET}/mobilenet-test
python3 main.py --tpu=$TPU_NAME --data_dir=$DATA_DIR --model_name=$MODEL_NAME \
--model_dir=$OUTPUT_DIR --export_dir=$OUTPUT_DIR/export
This repo's implementation is built upon the codebase of MnasNet's TPU implementation as well as that of Single-path NAS.