microsoft / vott Goto Github PK

also on export to CNTK and suggest tags

The load video sequence is a bit confusing – you shouldn’t be able to submit if you didn’t select a video

Add image support in addition to video tagging

Instead of exporting all frames that have no tags only export frames that people have visited

ipcMain.on('setWindowSize') stretches the tagging canvas and breaks the tagging mechanism.

A. auto runs tracking till the end of a scene
B. if the tracker is unsure about a suggestion prompts for user input
C. continues tracking to the end of a scene

Since a certain user can have multiple version of CNTK installed, we'll probably want to allow the user to choose with which version of CNTK he wants to work.
For now - we always assumes that the directory is named c:\local\cntk.
This requires a minor fix in the node-cntk-fastrcnn module, but the majority of the work should be on the UI.

I'm setting priority to P2, although it's a pretty critical feature in order to have the auto tagging useful IMO.

Repro steps:

• Open brand new video (no saved metadata)
• Tag a few frames
• Save video
• Open same video
• Notice that first tag isn't shown
• Step frame forward
• Notice that 2nd tag is shown
• Step frame backward
• Notice that 1st tag is shown (but wasn't when video was initially loaded)

looks like snapToAspectRatio might have broken this

@nadavbar need clarification

Update Tracking Blacklist to blacklist suggesting frame/region and not the suggested region.

Currently the tracking blacklist blacklists suggestions on next frame instead of the current frame updating this will make keeping track of suggestions much more effecient

Eventhough Frame Changes. The canvas is not loaded. Solution the onupdate event to canplaythrough.

repro:

• open file dialog
• while dialog is open, resize window
• open a video, select continue
• black screen

If i tag an object in a video preform blob tracking of the tagged div on the subsequent frames and provide suggested tags from that.

This thread discuss possible ways to reduce the package size: electron/electron#2003

Create a function that goes frame by frame into a function that takes a call back to clean the export to and review cntk code

Make the label default to the last used label instead of none or make it at least default to the first label if their is only one label provided instead of none.

Add a control that allows the user to quickly skip to any frame.

-Add a hover to visualize a frame on the skip bar

Add support for rendering the results of a model in the tool for suggested tags

A command line function to train a model using the "cntk export folder" that is generated by the tool.

1/2 day @nadavbar need clarification on which frames goes into which folder

Add an overlay and animation on top of the video during the export process

Can input 0 (which crashes the app after frame is stepped) or > 999

Add code to enable the training of a model from multiple tagged videos

A one time operation to to export frame of a video into a folder.

Loads metadata from batch cntk tagging

Lets make sure this is the most elegant way of doing this. I want to minimize the amount of forms and dialogs

The current implementation creates a new background video and iterator for every tag, it should only iterate once.

positive export files are missing after the export operation.