can this be converted to tflite model?

[farazBhatti]

No description provided.

[Akhilesh64]

I am not sure but tensorflow provides utilities to convert a model to tflite. Otherwise you can retrain the model and save it as a tflite model.

[farazBhatti]

here is the conversion code

import os
from keras.models import load_model
import tensorflow as tf

model = load_model('models/model_DUTS.h5', compile=False)

converter = tf.lite.TFLiteConverter.from_keras_model(model)

tfmodel = converter.convert()
open ("model.tflite" , "wb") .write(tfmodel)
print("Sucess!")

and this is for getting inference from converted tflite model

import tensorflow as tf
import PIL.Image as Image
import numpy as np
from sklearn.preprocessing import binarize
from tensorflow.keras.preprocessing.image import array_to_img

import cv2


def preprocess(img_path, dim):
    img = Image.open(img_path)
    img = img.resize(dim, Image.BILINEAR)
    img = np.array(img).astype(np.float32)
    img = np.expand_dims(img, axis=0)

    return img


w = 256
h = 256
dim = (w,h)
model = "model.tflite"
img_path = 'images/boat.jpg'


# Load the TFLite model and allocate tensors.
interpreter = tf.lite.Interpreter(model)
interpreter.allocate_tensors()

# Get input and output tensors.
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()


#Preprocess the image to required size and cast

img = preprocess(img_path, dim)
input_shape = input_details[0]['shape']

input_tensor= np.array(np.expand_dims(img,0))
input_tensor = input_tensor.reshape(input_shape)

#set the tensor to point to the input data to be inferred
input_index = interpreter.get_input_details()[0]["index"]
interpreter.set_tensor(input_index, input_tensor)

#Run the inference
interpreter.invoke()
preds = interpreter.get_tensor(output_details[0]['index'])

preds = np.squeeze(preds)

#For thresholding
for i in range(len(preds)):
    shape = preds[i,:,:].shape
    frame = binarize(preds[1,:,:], threshold = 0.5)
    frame = np.reshape(frame,(shape[0], shape[1]))

#For saving all the frames
for i in range(len(preds)):
  img = np.expand_dims(np.array(preds[i,:,:]),axis=-1)
  img = array_to_img(img)
  img.save('img'+str(i)+'.png')

[Akhilesh64]

I dont have any experience with tflite models. As for the prediction script make changes to the below script to suit your needs :

import cv2
import numpy as np
from keras.models import load_model
from sklearn.preprocessing import binarize
from tensorflow.keras.preprocessing.image import array_to_img

model = load_model('models/model_DUTS.h5', compile=False)

img = cv2.imread('images/spawn.png', cv2.IMREAD_COLOR)
img = cv2.resize(img, (256, 256))
img = np.expand_dims(img, axis=0)

preds = model.predict(img)
preds = np.squeeze(preds)

# For thresholding
for i in range(len(preds)):
    shape = preds[i, :, :].shape
    frame = binarize(preds[1, :, :], threshold=0.5)
    frame = np.reshape(frame, (shape[0], shape[1]))
    frame = np.expand_dims(np.array(preds[i,:,:]),axis=-1)
    frame = array_to_img(frame)
    frame.save('img'+str(i)+'.png')