Skip to content

Commit

Permalink
Balloon Color Splash sample.
Browse files Browse the repository at this point in the history
  • Loading branch information
waleedka committed Mar 19, 2018
1 parent 22a80f3 commit 555126e
Show file tree
Hide file tree
Showing 4 changed files with 2,578 additions and 0 deletions.
48 changes: 48 additions & 0 deletions samples/balloon/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,48 @@
# Color Spash Example

This is an example showing the use of Mask RCNN in a real application.
We train the model to detect balloons only, and then we use the generated
masks to keep balloons in color while changing the rest of the image to
grayscale.

## Installation
From the [Releases page](https://github.com/matterport/Mask_RCNN/releases) page:
1. Download `mask_rcnn_balloon.h5`. Save it in the root directory of the repo (the `mask_rcnn` directory).
2. Download `balloon_dataset.p3`. Expand it such that it's in the path `mask_rcnn/datasets/balloon/`.

## Apply color splash using the provided weights
Apply splash effect on an image:

```bash
python3 balloon.py splash --weights=/path/to/mask_rcnn/mask_rcnn_balloon.h5 --image=<file name or URL>
```

Apply splash effect on a video. Requires OpenCV 3.2+:

```bash
python3 balloon.py splash --weights=/path/to/mask_rcnn/mask_rcnn_balloon.h5 --video=<file name or URL>
```


## Run Jupyter notebooks
Open the `inspect_balloon_data.ipynb` or `inspect_balloon_model.ipynb` Jupter notebooks. You can use these notebooks to explore the dataset and run through the detection pipelie step by step.

## Train the Balloon model

Train a new model starting from pre-trained COCO weights
```
python3 balloon.py train --dataset=/path/to/balloon/dataset --weights=coco
```

Resume training a model that you had trained earlier
```
python3 balloon.py train --dataset=/path/to/balloon/dataset --weights=last
```

Train a new model starting from ImageNet weights
```
python3 balloon.py train --dataset=/path/to/balloon/dataset --weights=imagenet
```

The code in `balloon.py` is set to train for 3K steps (30 epochs of 100 steps each), and using a batch size of 2.
Update the schedule to fit your needs.
Loading

0 comments on commit 555126e

Please sign in to comment.