Migrating code to work on tensorflow 2.0 and adding running on directory #21
Conversation
| from keras.models import Model as KerasModel | ||
| from keras.layers import Input, Dense, Flatten, Conv2D, MaxPooling2D, BatchNormalization, Dropout, Reshape, Concatenate, LeakyReLU | ||
| from keras.optimizers import Adam | ||
|
|
| from pipeline import * | ||
|
|
||
| from keras.preprocessing.image import ImageDataGenerator | ||
| import os |
| # 3 - Predict | ||
| X, y = generator.next() | ||
| print('Predicted :', classifier.predict(X), '\nReal class :', y) | ||
| num_iterations = 0 |
There was a problem hiding this comment.
For this change in the predict file, I don't really want to provide a loop, this was a minimal working code to show that this uses the tf/keras framework, but there are many ways to loop over the images. So I would leave this file unchanged while letting the new script predict_on_directory do want you want to do.
| @@ -0,0 +1,35 @@ | |||
| ######################################################################################################################## | |||
| # Model | |||
| ######################################################################################################################## | |||
There was a problem hiding this comment.
inaccurate comment, delete, (or change and use triple quotes to provide a documentation on the command line usage)
|
|
||
| # 1 - Load the model and its pretrained weights | ||
| classifier = Meso4() | ||
| classifier.load('weights/Meso4_DF') |
There was a problem hiding this comment.
Well, for the purpose of a command line script, this needs parametrisation, what if your directory is forged using face-2-face.
| im = load_img(f, target_size=REQUIRED_SIZE) | ||
| im_arr = np.expand_dims(img_to_array(im), axis=0) | ||
| im_arr /= 255.0 | ||
| print(im_arr.shape) |
| # Getting files | ||
| files = glob.glob(os.path.join(images_dir, "*.jpg")) | ||
| for f in files: | ||
| im = load_img(f, target_size=REQUIRED_SIZE) |
There was a problem hiding this comment.
it's better if you wrap the image loader into a ImageDataGenerator and use flow_from_directory than manually loading the images
|
Thanks for your contribution and interest for this work, a command line script is indeed a good addition to this repo. I've commented several little thing on your proposed script. |
3 participants
Migrating code to work on tensorflow 2.0 and adding running on directory