Open showkeyjar opened 2 years ago
I fig it out, would you please check this idea is right?
I rewrite a OcclusionSensitivity class and use python read tf.lite model then predict
the key codes:
` def get_sensitivity_map(self, interpreter, image, class_index, patch_size):
...
interpreter.allocate_tensors()
for patch in tqdm(patches):
interpreter.set_tensor(interpreter.get_input_details()[0]['index'], [patch])
interpreter.invoke()
preds = interpreter.get_tensor(interpreter.get_output_details()[0]['index'])
predictions.extend(preds)
target_class_predictions = [
prediction[class_index] for prediction in predictions
]
...
`
beacause I can't find how to batch predict image use tf.lite model, so I make a for loop and set a bigger patch_size
can tf-expain explain tf-lite models?
if can, how to explain?
if not, would you have some advice or suggestion?