Skip to content

Conversation

@kartikdutt18
Copy link
Member

@kartikdutt18 kartikdutt18 commented Jun 5, 2020

This is nearly the same PR as the other image dataloader. Difference is that I have used field vector inserting bounding boxes.
So in preprocessing of dataset for model we would have instant lookups.

@zoq
Copy link
Member

zoq commented Jun 5, 2020

This is nearly the same PR as the other image dataloader. Difference lies that I have used field vector inserting bounding boxes.
So in preprocessing of dataset for model we would have instant lookups.

So I guess we can close the other one?

@KimSangYeon-DGU
Copy link
Member

KimSangYeon-DGU commented Jun 5, 2020

@zoq Yes, we can close the other one if this is better :) We're currently seeing if which approach is better between this and #13.

Copy link
Member

@KimSangYeon-DGU KimSangYeon-DGU left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Left some style and possible improvement comments, it looks good to go :)

Copy link
Member

@KimSangYeon-DGU KimSangYeon-DGU left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Another comment

Copy link
Member

@saksham189 saksham189 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just some minor comments. Looks great!

@kartikdutt18 kartikdutt18 force-pushed the ImageDataloadersBetter branch from 5b8e0c9 to 1218f53 Compare June 8, 2020 08:24
Copy link
Member

@KimSangYeon-DGU KimSangYeon-DGU left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it's almost ready to be merged, please check the comments including the source path of serverName in the CurlDownloadTest. I mean this.

Copy link
Member

@zoq zoq left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Really nice to see this come to live, do you think we could provide a simple notebook to show how this code can be used to augment some image dataset?

@kartikdutt18
Copy link
Member Author

kartikdutt18 commented Jun 10, 2020

do you think we could provide a simple notebook to show how this code can be used to augment some image dataset?

Sure I can give it a try, This would be for examples repo or model repo? Thanks.

Copy link
Member

@KimSangYeon-DGU KimSangYeon-DGU left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks great! I left some minor comments :)

@kartikdutt18 kartikdutt18 force-pushed the ImageDataloadersBetter branch from b0a8c2c to 022ba45 Compare June 11, 2020 06:00
@KimSangYeon-DGU
Copy link
Member

@kartikdutt18 I think this is ready to be merged. Can you squash the commits to be concise?

@kartikdutt18
Copy link
Member Author

Sure. Let me do that right now.

Trained a lenet1 model

Add All Weights

Add unzip utility function, fix linux build

Completed everything

Fix build, Style fixes next

Use force local for windows while unzipping tar files

Use force local for windows while unzipping tar files

Added Utility Function

Added Utility Function

Style Fix

Fix Typo causing build error in windows

Fix Typo causing build error in windows

Fix const issue for windows

Extract in data folder

Reposition force local

This should work

Print Path in windows for debugging

Print Path in windows for debugging

Print Path in windows for debugging

Stip components of tar

Stip components of tar

strip componenets

Initial Definition

Add List Dir utility function

Add List Dir utility function

Squash this, annotation read

Yay, we can read xml now

Complete labels portion of dataloader

Style Fix

Style Fix

Style Fix

Add basic definition of augmentation class

Load images according to annotation in column format, Reduce complexity by preventing whole iteration

Augmentation work

Augmentation work
Style fixes

Style fixes
@kartikdutt18 kartikdutt18 force-pushed the ImageDataloadersBetter branch from 022ba45 to c859e8e Compare June 11, 2020 06:31
Copy link
Member

@KimSangYeon-DGU KimSangYeon-DGU left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Amazing! Looks good to me 👍

@kartikdutt18
Copy link
Member Author

Pushing a commit to delete the dataset.

Upload dataset temp

Resolve merge conflicts

Resolve merge conflicts

Resolve merge conflicts

Resolve merge conflicts

Style fixes

Set up constructor

Complete Dataloader, write function definition for image dataloader

Build Fixed, Style Fixed

style fixed

Fix linux and Windows build by ignoring invalid files like .DS_Store

Add image dataloader, needs clean up

Style fixes

Check for file type

Check for correct extention and filetype

Style Fixed

Build Fixed, Nice way to avoid hidden files

Style Fixed

Clean Up

Why is config there even though I did a rebase

Add data split (internal), need implementation only for field type

Style Fixed

need to seperate the declaration

Fixed

style fixes, commits also need clean up

Remove template from Augmentation

Fixed the duplicacy error

Add mat type support (there is invalid read)

Mat type gives invalid read

Hmm,

style fixes, commits also need clean up

Typo -> invalid read -> Fixed

Remove extra lines

Rename cell to image

Style Fixes and use mlpack URL

Make augmentation case insensitive

Allow multiple Resize Transform

Add unknown augmentation warning

Style Fixes and boundary checks

Increase dataset size, adjust comments, change URL

Style Fixes and boundary checks

Remove locally held datasets
@kartikdutt18 kartikdutt18 force-pushed the ImageDataloadersBetter branch from c859e8e to 3fc6e7b Compare June 11, 2020 12:29
Copy link
Member

@saksham189 saksham189 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the work on this! LGTM

@zoq
Copy link
Member

zoq commented Jun 11, 2020

do you think we could provide a simple notebook to show how this code can be used to augment some image dataset?

Sure I can give it a try, This would be for examples repo or model repo? Thanks.

I think it makes sense to put it into the examples repository, super excited to see this merged.

@KimSangYeon-DGU KimSangYeon-DGU merged commit 5ea20fc into mlpack:master Jun 11, 2020
@KimSangYeon-DGU
Copy link
Member

@kartikdutt18 Thanks for the amazing work!

@kartikdutt18
Copy link
Member Author

Thanks a lot @KimSangYeon-DGU, @zoq, @saksham189 for the reviews and all the help.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants