efficientnet_lite

EfficientNetLite

Input

Ailia input shape: (224, 224, 3)
Range: [-128, 127] 8-bit signed integer

Output

class_count=3
+ idx=0
  category=409[analog clock ]
  prob=5
+ idx=1
  category=892[wall clock ]
  prob=-52
+ idx=2
  category=826[stopwatch, stop watch ]
  prob=-119
Script finished successfully.

Usage

Automatically downloads the tflite file on the first run. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 efficientnet_lite.py 

If you want to specify the input image, put the image path after the --input option.
You can use --savepath option to change the name of the output file to save.

$ python3 efficientnet_lite.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 efficientnet_lite.py --video VIDEO_PATH --savepath SAVE_VIDEO_PATH

Reference

TensorFlow model

• efficientnet-lite-keras

Torch model

• pytorch-image-models

Framework

TensorFlow 2.6, Torch 1.13.0

Netron

TensorFlow model

• efficientnetliteb0_quant.tflite
• efficientnetliteb0_quant_recalib.tflite
• efficientnetliteb0_float.tflite

Torch model

• efficientnetliteb0_torch_quant.tflite
• efficientnetliteb0_torch_float.tflite