Bug fixes and minor refactoring #2
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…fix incompatibility in newer versions
…arse to fix a bug where arguments weren't taken into account
starkgate
commented
Sep 10, 2020
| ### Dockerfile used for augmentation pipeline code | ||
| ### | ||
| FROM gcr.io/tensorflow/tensorflow:latest-gpu | ||
| FROM tensorflow/tensorflow:1.15.2-gpu |
Author
There was a problem hiding this comment.
gcr.io doesn't host tensorflow dockers anymore
starkgate
commented
Sep 10, 2020
| FROM tensorflow/tensorflow:1.15.2-gpu | ||
| MAINTAINER Vincent Vanhoucke <vanhoucke@google.com> | ||
| RUN pip install scikit-learn scikit-image opencv-python==3.2.0.8 | ||
| RUN pip install scipy==1.1.0 scikit-learn scikit-image opencv-python==3.2.0.8 |
Author
There was a problem hiding this comment.
scipy 1.2.0+ removes scipy.imread and probably other functions needed by SEA. It was faster to install an older version than refactor the code for newer scipy
starkgate
commented
Sep 10, 2020
| # | ||
| # generate the CFA arrays for R,G,B based upon the r pixel location: | ||
| # | ||
| h = int(im_h / 2) |
Author
There was a problem hiding this comment.
Leaving h and w as floats causes a crash in np.tile
starkgate
commented
Sep 10, 2020
Comment on lines
-9
to
+23
| flags = tf.app.flags | ||
| flags.DEFINE_integer("epoch", 1, "number of epochs; corresponds to number of augmentations to perform on the dataset (i.e., epoch =2 means the dataset will be augmented twice") | ||
| flags.DEFINE_integer("batch_size", 2, "The size of batch images; must be a multiple of n and >1") | ||
| flags.DEFINE_integer("c_dim", 3, "Dimension of image color. [3]") | ||
| # | ||
| flags.DEFINE_string("Img_dataset","generator_images","The name (full path) of dataset to augment") | ||
| flags.DEFINE_integer("Img_height",512, "The size of the output images to produce [64]") | ||
| flags.DEFINE_integer("Img_width", 1024, "The size of the output images to produce. If None, same value as output_height [None]") | ||
| flags.DEFINE_boolean("chromab_flag", True, "flag that specifies whether to perform Chromatic aberration augmentation") | ||
| flags.DEFINE_boolean("blur_flag", True, "flag that specifies whether to perform Blur augmentation") | ||
| flags.DEFINE_boolean("exposure_flag", True, "flag that specifies whether to perform Exposure augmentation") | ||
| flags.DEFINE_boolean("noise_flag", True, "flag that specifies whether to perform noise augmentation") | ||
| flags.DEFINE_boolean("color_flag", True, "flag that specifies whether to perform color shift augmentation") | ||
| flags.DEFINE_boolean("save_aug_params_flag", False, "flag that specifies whether to save aug. parameters for each image") | ||
| # | ||
| flags.DEFINE_string("input_fname_pattern", "*.png", "Glob pattern of filename of input images [*]") | ||
| flags.DEFINE_string("results_dir", "results", "Directory name to save the augmented images [results]") | ||
| FLAGS = flags.FLAGS | ||
| parser = argparse.ArgumentParser(description='Augment a dataset') | ||
| parser.add_argument('-n', type=int, default=1, nargs='?', help='sets the number of augmentations to perform on the dataset i.e., setting n to 2 means the dataset will be augmented twice') | ||
| parser.add_argument('-b', '--batch_size', type=int, default=64, nargs='?', help='size of batches; must be a multiple of n and >1') | ||
| parser.add_argument('-c', '--channels', type=int, default=3, nargs='?', help='dimension of image color channel (note that any channel >3 will be discarded') | ||
| parser.add_argument('-i', '--input', type=str, help='path to the dataset to augment') | ||
| parser.add_argument('-o', '--output', type=str, default='results', nargs='?', help='path where the augmented dataset will be saved') | ||
| parser.add_argument('--pattern', type=str, default="*.png", nargs='?', help='glob pattern of filename of input images') | ||
| parser.add_argument('--image_height', type=int, default=512, nargs='?', help='size of the output images to produce (note that all images will be resized to the specified image_height x image_width)') | ||
| parser.add_argument('--image_width', type=int, default=1024, nargs='?', help='size of the output images to produce. If None, same value as output_height') | ||
| parser.add_argument('--chromatic_aberration', type=bool, default=False, nargs='?', help='perform chromatic aberration augmentation') | ||
| parser.add_argument('--blur', type=bool, default=False, nargs='?', help='perform blur augmentation') | ||
| parser.add_argument('--exposure', type=bool, default=False, nargs='?', help='perform exposure augmentation') | ||
| parser.add_argument('--noise', type=bool, default=False, nargs='?', help='perform noise augmentation') | ||
| parser.add_argument('--colour_shift', type=bool, default=False, nargs='?', help='perform colour shift augmentation') | ||
| parser.add_argument('--save_params', type=bool, default=False, nargs='?', help='save augmentation parameters for each image') | ||
| args = parser.parse_args() |
Author
There was a problem hiding this comment.
Replaced tf's deprecated flags with the much more robust argparse. Cleaned up the arguments to hopefully be clearer
starkgate
commented
Sep 10, 2020
| input_fname_pattern='*.png', results_dir=None): | ||
| def __init__(self, sess, image_height, image_width, batch_size, channels, | ||
| input, colour_shift, chromatic_aberration, blur, exposure, noise, save_params, | ||
| pattern, output): |
Author
There was a problem hiding this comment.
Defaults aren't required with argparse
starkgate
commented
Sep 10, 2020
| ## construct the graph of the image augmentation architecture | ||
| ## | ||
| print 'building model/graph' | ||
| print('building model/graph') |
Author
There was a problem hiding this comment.
Python 3 requires parentheses around print
starkgate
commented
Sep 10, 2020
| NoiseParams = np.array(out[0][4]) | ||
| ColorParams = np.array(out[0][5]) | ||
| ## save images | ||
| self.save_augmented_final_images(G_output_images, G_batch_files, ChromAbParams, BlurParams, ExpParams, NoiseParams, ColorParams, n) |
Author
There was a problem hiding this comment.
This if/else isn't necessary, the check for save_params is already done is save_augmented_final_images
starkgate
commented
Sep 10, 2020
| code_location="/home/u42/Windows/Data/datasets/SensorEffectAugmentation" | ||
| input="/home/u42/Documents/mmdetection/data/synth_rocks/train/" | ||
| output="/home/u42/Documents/mmdetection/data/synth_rocks/train_augmented/" | ||
|
|
starkgate
commented
Sep 10, 2020
| from time import gmtime, strftime | ||
|
|
||
| pp = pprint.PrettyPrinter() | ||
|
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
On top of issue #1, I encountered a few bugs and crashes when using SensorEffectAugmentation (SEA), which I fixed in these commits. The issues are probably caused by newer versions of the various tools and libraries used by SEA. Either way, I hope this is useful!