Introduces a Random resized crop layer #203

frostbyte012 · 2022-03-24T09:11:49Z

Feature Implementation : Random Resized Crop Layer (Standard Augmentation Transformation).
References issue: #131

…f the images in batches

keras_cv/layers/preprocessing/random_resized_crop.py

LukeWood · 2022-03-24T18:05:09Z

keras_cv/layers/preprocessing/random_resized_crop.py

+
+    # method for resizing the images
+    def _resize(self, inputs):
+        outputs = image_utils.smart_resize(inputs, [self.height, self.width])


This will perform the same resize on all images

Done : called on import random
and used it to randomly resize height and width of these images, for further implementation of random crop.

LukeWood · 2022-03-24T18:06:39Z

Thanks for the PR! Ideally the layer would:

randomly resize each image individually
randomly crop each image individually from those resizes

The resulting images would have the same size, but different scalings

frostbyte012 · 2022-03-26T21:24:20Z

Welcome!
I figured out the problems.
Trying to fix it and resolve the changes as soon as possible.

…Crop

LukeWood · 2022-03-28T03:28:32Z

keras_cv/layers/preprocessing/random_resized_crop.py

+W_AXIS = -2
+
+
+class RandomResizedCrop(base_layer.Layer):


lets subclass tf.keras.internal.layers.BaseImageAugmentationLayer instead of base_layer.Layer.

Check out
https://github.com/keras-team/keras-cv/blob/master/keras_cv/layers/preprocessing/random_shear.py

Done : Added a subclass tf.keras.internal.layers.BaseImageAugmentation.Layer in place of base_layer.Layer

LukeWood · 2022-03-28T03:29:07Z

keras_cv/layers/preprocessing/random_resized_crop.py

@@ -0,0 +1,91 @@
+from keras import backend


lets also add a script to examples showing the outputs of this layer. Lets base it on the color degeneration example:

https://github.com/keras-team/keras-cv/blob/master/examples/layers/preprocessing/random_color_degeneration_demo.py

LukeWood · 2022-03-28T03:29:25Z

Thanks for the updates!

LukeWood · 2022-03-28T19:46:53Z

keras_cv/layers/preprocessing/random_resized_crop.py

+
+    # method for resizing the images
+    def _resize(self, inputs):
+        self.new_height=random.randint(0,self.height)


so you'll actually need to use a tf.random and use the get_random_transformation API in the BaseImageAugmentation layer. This random value gets baked into the computation graph

Done : Resolved it and modified it using the tf.random and get_random_transformation API from BaseImageAugmentation layer

frostbyte012 · 2022-03-30T20:04:38Z

I reviewed the changes that are suggested and I'm starting to work on them as soon as possible.

…yer istead of base_layer.Layer

LukeWood · 2022-04-01T17:14:45Z

I reviewed the changes that are suggested and I'm starting to work on them as soon as possible.

Great! Thanks for your hard work @frostbyte012

innat · 2022-04-16T20:23:48Z

@LukeWood cc. @qlzh727
The RandomBrightness was moved to core Keras instead of Keras-cv, described here. In the same way, isn't it more suitable to have this layer in core Keras? As in core keras, we have resizing, and cropping!

LukeWood · 2022-04-21T17:33:31Z

Hey @frostbyte012 are you planning to continue work on this feature? If not, I may begin working on this. This feature will (at some point soon) become a priority, as the layer is very important to the ecosystem!

Let me know! No pressure, but want to make sure we can move forward on this soon.

frostbyte012 · 2022-04-21T17:38:32Z

@LukeWood yes sir, I'm extremely sorry I wasn't well for few days , but I'm good to work on with it now, I really am passionate and excited enough to work on with this layer. I'm going to work on this from tomorrow onwards. I'm extremely sorry once again I should have informed about it to you before.

LukeWood · 2022-04-21T17:41:40Z

@LukeWood yes sir, I'm extremely sorry I wasn't well for few days , but I'm good to work on with it now, I really am passionate and excited enough to work on with this layer. I'm going to work on this from tomorrow onwards. I'm extremely sorry once again I should have informed about it to you before.

Absolutely no worries - being an OSS contributor should be enjoyable and not stressful. Just wanted to check in. Thanks!

frostbyte012 · 2022-04-22T16:33:36Z

I'm working on the changes requested asap! Will be giving the updates as fast as possible

LukeWood · 2022-04-22T17:30:07Z

keras_cv/layers/preprocessing/random_resized_crop.py

+        # Attribute to take random seed
+
+    #Methoid for taking the inputs
+    def call(self, inputs, training=True):


You'll need to override augment_image instead of call when subclassing BaseImageAugmentationLayer

sure sir working on it Asap!

Done: Resolved it and applied augment_image

@LukeWood sir is the last commit all, correct? If not please let me know about it! Waiting to make the necessary changes. if required.

…seImageAugmentation layer.

…entationLayer.

innat · 2022-04-26T16:05:43Z

keras_cv/layers/preprocessing/random_resized_crop.py

+import numpy as np
+import tensorflow as tf
+from tensorflow.python.util.tf_export import keras_export
+from tensorflow.tools.docs import doc_controls


@frostbyte012

keras should be imported from tensorflow, for example: from tensorflow.keras import ....

remove not-used packages, for ex. keras_export.

Agreed! Thanks @innat

Sure @LukeWood and @innat sir! I'm removing the unwanted packages and also adding the correct format for keras import .

keras_cv/layers/preprocessing/random_resized_crop.py

innat · 2022-04-26T16:08:24Z

keras_cv/layers/preprocessing/random_resized_crop.py

+        self.width = width
+        # Attribute to take the custom width
+        self.seed = seed
+        # Attribute to take random seed


It may not be needed to give obvious comments.

innat · 2022-04-26T16:14:44Z

@frostbyte012 Great work.
Is it possible to share some demo output of this layer?

frostbyte012 · 2022-04-26T19:15:34Z

@innat sure sir! Thanks!
Yes sir I'll most probably add an example of this layer to the examples

LukeWood

Thanks for the PR! I left a few comments, but there are some fundamental changes that should occur. My guide on using BaseImageAugmentationLayer will cover these.

I'll link you to the guide when it is ready.

LukeWood · 2022-04-26T23:45:29Z

keras_cv/layers/preprocessing/random_resized_crop.py

+import numpy as np
+import tensorflow as tf
+from tensorflow.python.util.tf_export import keras_export
+from tensorflow.tools.docs import doc_controls


Agreed! Thanks @innat

keras_cv/layers/preprocessing/random_resized_crop.py

frostbyte012 · 2022-04-28T20:02:02Z

@LukeWood and @innat sir I applied the suggested changes and working on the examples soon !
If you have some other changes or suggestions kindly let me know.
Thanks for the reviews!

…o random_resized_crop_branch

frostbyte012 · 2022-04-30T20:13:04Z

@innat and @LukeWood sir ! I just added a demo template test for random resized crop taking the random shear layer reference from preprocessing layers, kindly check the demo template and help me adjust and make the necessary changes, If require so. Thanks!

innat · 2022-05-01T07:53:11Z

@frostbyte012 I was asking if you could display some sample output like this.

innat · 2022-05-01T08:03:56Z

keras_cv/layers/preprocessing/random_resized_crop.py

+from keras.utils import image_utils
+
+@tf.keras.utils.register_keras_serializable(package="keras_cv")
+class RandomResizedCrop(tf.keras.internal.layers.BaseImageAugmentationLayer):


I think it should be __internal__.

LukeWood · 2022-05-05T17:20:10Z

Hey @frostbyte012, make sure to run unit tests before requesting another review! Really glad you are excited to contribute, but reviewer time is quite limited and as such it is best to be sure to test all code before requesting review.
E.x.: tf.keras.internal will fail to run as it should be .internal, and augment_image must augment a single image instead of a batch, etc.

Thanks @frostbyte012 and I'm glad you're excited to contribute!

frostbyte012 · 2022-05-06T19:55:12Z

@LukeWood I'm extremely sorry sir! I'm making sure to implement the changes and try on the unit test and make sure to get along with it ASAP! Extremely sorry for the inconvenience caused!

LukeWood · 2022-05-12T18:08:04Z

Hey @frostbyte012! This item is pretty high on our critical roadmap, and I think we will need to pick it up to carry forward. I'm really glad you're excited to contribute to the repo! Maybe a great starting point would be this issue instead?

#425

Thanks!

frostbyte012 · 2022-05-18T21:51:27Z

I'm extremely sorry @LukeWood @innat couldn't keep up my promise and couldn't fulfill my commitment, I'm kinda sad about it, anyways thanks for reconsidering me for working with the other issue, I'll definitely not give up.. failing 100 times but rising the 101th time that should be our motto, I may not be experienced and not talented, but I wish my passion and dedication outperforms that. I was having some issues but now I'm comfortable, I wish to start working soon. Thanks

LukeWood · 2022-05-18T21:55:36Z

I'm extremely sorry @LukeWood @innat couldn't keep up my promise and couldn't fulfill my commitment, I'm kinda sad about it, anyways thanks for reconsidering me for working with the other issue, I'll definitely not give up.. failing 100 times but rising the 101th time that should be our motto, I may not be experienced and not talented, but I wish my passion and dedication outperforms that. I was having some issues but now I'm comfortable, I wish to start working soon. Thanks

Do not worry at all! This is actually a very tricky issue. I tried to work on it the other day, and there are a LOT of edge cases to consider.

By the way, the guide on BaseImageAugmentationLayer is now live.

https://keras.io/guides/keras_cv/custom_image_augmentations/

I will take on RandomlyResizedCrop as it is really tricky! Please keep at it and I am sure we will get a contribution merged! I appreciate your hard work.

Also tweaked documentation of `Cropping1D` for consistency.

frostbyte012 added 2 commits March 22, 2022 20:28

build : Add random_resized_crop.py to implement RandomResizedCrop Layer

bb442c9

build : Add a new random resized crop layer for the random resizing o…

e19e665

…f the images in batches

LukeWood suggested changes Mar 24, 2022

View reviewed changes

fix : Made the necessary changes for Implementing the Random Resized …

56680d5

…Crop

LukeWood reviewed Mar 28, 2022

View reviewed changes

LukeWood suggested changes Mar 28, 2022

View reviewed changes

LukeWood changed the title ~~Random resized crop branch~~ Introduces a Random resized crop layer Mar 28, 2022

LukeWood reviewed Mar 28, 2022

View reviewed changes

fix : Add a subclass tf.keras.internal.layers.BaseImageAugmentationLa…

113b6f7

…yer istead of base_layer.Layer

build : Add subclass tf.keras.internal.layers.BaseImageAugmentationLayer

a7373b8

LukeWood reviewed Apr 22, 2022

View reviewed changes

frostbyte012 added 2 commits April 23, 2022 01:25

fix : Use tf.random and applied get_random_transformation API from Ba…

fbf7642

…seImageAugmentation layer.

fix : Use augment_image instead of call for subclassing BaseImageAugm…

1eec92a

…entationLayer.

innat reviewed Apr 26, 2022

View reviewed changes

keras_cv/layers/preprocessing/random_resized_crop.py Outdated Show resolved Hide resolved

innat reviewed Apr 26, 2022

View reviewed changes

LukeWood suggested changes Apr 26, 2022

View reviewed changes

fix : use correct importing format for keras library

4bf8af7

frostbyte012 added 2 commits April 29, 2022 01:16

fix : remove obvious comments

9273df4

fix : remove the constants declared outside the class

be11a7c

frostbyte012 added 3 commits April 29, 2022 01:53

fix : remove the H_AXIS and W_AXIS outside the class

664fbf3

Merge branch 'master' of https://github.com/frostbyte012/keras-cv int…

330d4ec

…o random_resized_crop_branch

build : a demo framework for random_resized_crop made

9462864

innat reviewed May 1, 2022

View reviewed changes

LukeWood closed this May 12, 2022

freedomtan pushed a commit to freedomtan/keras-cv that referenced this pull request Jul 20, 2023

Add keras_core.layers.ZeroPadding1D. (keras-team#203)

ca39016

Also tweaked documentation of `Cropping1D` for consistency.

Introduces a Random resized crop layer #203

Introduces a Random resized crop layer #203

Conversation

frostbyte012 commented Mar 24, 2022

Choose a reason for hiding this comment

frostbyte012 Mar 27, 2022 • edited Loading

Choose a reason for hiding this comment

LukeWood commented Mar 24, 2022

frostbyte012 commented Mar 26, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

LukeWood commented Mar 28, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

frostbyte012 commented Mar 30, 2022

LukeWood commented Apr 1, 2022

innat commented Apr 16, 2022

LukeWood commented Apr 21, 2022

frostbyte012 commented Apr 21, 2022

LukeWood commented Apr 21, 2022

frostbyte012 commented Apr 22, 2022

Choose a reason for hiding this comment

Choose a reason for hiding this comment

frostbyte012 Apr 22, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

innat commented Apr 26, 2022

frostbyte012 commented Apr 26, 2022

LukeWood left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

frostbyte012 commented Apr 28, 2022

frostbyte012 commented Apr 30, 2022 • edited Loading

innat commented May 1, 2022

Choose a reason for hiding this comment

LukeWood commented May 5, 2022

frostbyte012 commented May 6, 2022

LukeWood commented May 12, 2022

frostbyte012 commented May 18, 2022

LukeWood commented May 18, 2022

frostbyte012 Mar 27, 2022 •

edited

Loading

frostbyte012 Apr 22, 2022 •

edited

Loading

frostbyte012 commented Apr 30, 2022 •

edited

Loading