-
Notifications
You must be signed in to change notification settings - Fork 1.4k
Object Detection walk-through #803
Object Detection walk-through #803
Conversation
|
||
### Loading the data into DIGITS | ||
|
||
To create an Object Detection dataset in DIGITS, [this version of `3.4-dev`](https://github.com/NVIDIA/DIGITS/tree/2613082eec6c207548c99bb69fa5defc9473b1e8), or any more recent version, is required. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Maybe just say DIGITS 4.0?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I tried to train detectNet with Digits. Everything is OK after I applied the latest changes but the result is always 0... 50 zero boxes after forward_net. I don't know why.
I used the prototxt in NVIDIA/caffe/examples/kitti. Do you have any suggestion?
The screenshots for the home page and for the "New Object Detection Dataset" page need updating with the latest UI changes. |
} | ||
} | ||
``` | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think that's mentioned on line 83
@gheinrich I will try it on Windows and get the results. |
3b9ff66
to
0c36fce
Compare
Isaac, if you do want to try this on Windows, check the latest updates to this pull request. I updated it to change the suggested learning rate, etc. |
UI has changed in some places Convert all images from PNG to JPEG for sake of space
0c36fce
to
a81b526
Compare
Hi, I was checking this walk-through to reproduce the results. => could you please provide the links (or the name of the zip files) to the images and labels to be used for this walk-through? Thanks |
It would definitely be helpful to more clearly define the datasets we're to download.
|
Download left color images of object data set (12 GB) I think these two sets are what you need. |
Added a link to the KITTI object detection webpage with directions as to which files are needed for the walk-through example. Thanks to douzsh for his help: NVIDIA#803 (comment)
…kthrough Object Detection walk-through
No description provided.