Detect objects in images right in your browser using Tensorflow.js! Currently takes ~800ms to analyze each frame on Chrome MBP 13" mid-2014.
Supports Tiny YOLO
, as of right now,
tfjs
does not have
support to run any full YOLO models (and your user's computers probably
can't handle it either).
(You can only get so far with 1 FPS)
yarn add tfjs-yolo-tiny
npm install tfjs-yolo-tiny
import yolo, { downloadModel } from 'tfjs-yolo-tiny';
const model = await downloadModel();
const inputImage = webcam.capture();
const boxes = await yolo(inputImage, model);
// Display detected boxes
boxes.forEach(box => {
const {
top, left, bottom, right, classProb, className,
} = box;
drawRect(left, top, right-left, bottom-top, `${className} ${classProb}`)
});
Param | Type | Default | Description |
---|---|---|---|
input | tf.Tensor | - | Expected shape (1, 416, 416, 3) Tensor representing input image (RGB 416x416) |
model | tf.Model | - | Tiny YOLO tf.Model |
[options] | Object | See Below | Optional, Additional Configs |
If you're using a custom Tiny YOLO model or want to adjust the default filtering cutoffs, you may do so by passing an additional options object.
Example: yolo(inputImage, model, { classProbThreshold: 0.8 });
Option | Type | Default | Description |
---|---|---|---|
[options.classProbThreshold] | Number |
0.4 |
Filter out classes below a certain threshold |
[options.iouThreshold] | Number |
0.4 |
Filter out boxes that have an IoU greater than this threadhold (refer to tf.image.nonMaxSuppression) |
[options.filterBoxesThreshold] | Number |
0.01 |
Threshold to filter out box confidence * class confidence |
[options.maxBoxes] | Number |
2048 |
Number of max boxes to return, refer to tf.image.nonMaxSuppression. Note: The model itself can only return so many boxes. |
[options.yoloAnchors] | tf.Tensor |
See src/postprocessing.js |
(Advanced) Yolo Anchor Boxes, only needed if retraining on a new dataset |
[options.width] | Number |
416 |
(Advanced) If your model's input width is not 416, only if you're using a custom model |
[options.height] | Number |
416 |
(Advanced) If your model's input height is not 416, only if you're using a custom model |
[options.numClasses] | Number |
80 |
(Advanced) If your model has a different number of classes, only if you're using a custom model |
[options.classNames] | Array.<String> |
See src/coco_classes.js |
(Advanced) If your model has non-MSCOCO class names, only if you're using a custom model |
Returns an array of objects.
Property | Type | Description |
---|---|---|
top | Number | Pixels from top of image where bounding box starts |
left | Number | Pixels from left of image where bounding box starts |
bottom | Number | Pixels from top of image where box ends. |
right | Number | Pixels from left of image where box ends. |
classProb | Number | Probability of the class in the bounding box. |
className | String | Human name of the class. |
Param | Type | Default | Description |
---|---|---|---|
url | string | See DEFAULT_MODEL_LOCATION | Tiny YOLO Model config path. See tf.loadModel |
Returns a Promise
that can resolve to a tf.Model
.
PR's are more than welcome! Perf improvement or better test coverage are probably the two biggest areas of immediate need. If you have thoughts on extensibility as well, feel free to open an issue!
yarn install
If you're running tests, make sure to yarn add @tensorflow/[email protected]
so that you you don't get tfjs package not found errors. If you're developing,
make sure to remove tfjs as a dependency, as it'll start using the
local version of tfjs
intead of the peer version.
Note: Test coverage is poor, definitely don't rely on them to catch your errors.
yarn test
yarn build
Or during development, use watch mode, you can use the demo app to test out changes.
yarn watch