Add instance segmentation model #3
Conversation
|
Things to do to improve the experience with BrainFrame:
|
|
|
||
| ### Sources | ||
| Trained using the [Tensorflow Object Detection API]( | ||
| https://github.com/tensorflow/models/blob/master/research/object_detection) |
There was a problem hiding this comment.
Nit: use consistent link syntax
| ) | ||
|
|
||
| @staticmethod | ||
| def segm_postprocess(box, raw_cls_mask, im_h, im_w): |
| return im_mask | ||
|
|
||
| @staticmethod | ||
| def expand_box(box, scale): |
| **common_detector_options, | ||
| "keep_aspect_ratio": BoolOption( | ||
| default=False, | ||
| description="If True, it will keep the aspect ratio of " |
There was a problem hiding this comment.
Nit: I'm not sure of the technical term for this, but don't use "it will". The "it" can lead to confusion.
ex:
"If True, it will keep the aspect ratio" -> "If True, the aspect ratio will be kept"
| @@ -0,0 +1,13 @@ | |||
| Copyright 2020 Dilili Labs, Inc. | |||
There was a problem hiding this comment.
I thought we were using Aotu in the License files from now on?
| @@ -0,0 +1,201 @@ | |||
| Apache License | |||
There was a problem hiding this comment.
seems that this file is a duplicate, just with a lowercase name
| @@ -0,0 +1,2 @@ | |||
| [about] | |||
| api_compatibility_version = 0.2 | |||
There was a problem hiding this comment.
Nit: newline at end of file
|
|
||
| ## Model | ||
| ### Architecture & Performance | ||
| This model is is a Mask R-CNN with ResNet50 backbone. |
| this [python demo][python demo]. | ||
|
|
||
| ### Model File Origin | ||
| The pretrained model was downloaded from the Open |
There was a problem hiding this comment.
"Open OpenCV Model Zoo"
was this what you wanted it to say?
Postgres has a native Polygon type. We may want to use it: |
| @@ -0,0 +1,13 @@ | |||
| Copyright 2020 Dilili Labs, Inc. | |||
There was a problem hiding this comment.
File duplicate, but with lowercase name?
| state: BaseStreamState) -> DETECTION_NODE_TYPE: | ||
| frame_h, frame_w, _ = frame.shape | ||
|
|
||
| if options["keep_aspect_ratio"]: |
There was a problem hiding this comment.
Do you think people will want to turn this off? In other capsules, we always keep the same aspect ratio, right?
| @@ -0,0 +1,201 @@ | |||
| Apache License | |||
There was a problem hiding this comment.
We're not consistent about this, but the license should include the copyright owner as well, which for this model is "Copyright (c) 2019 Intel Corporation".
|
Postgres is so cool! BF-1440 |
| ) | ||
| else: | ||
| zipped = zip(boxes, classes, prediction['raw_masks'], scores) | ||
| for box, cls, raw_mask, score in zipped: |
There was a problem hiding this comment.
Change cls to class_id to match the pattern above
I'm creating this pull request to add an instance segmentation model to our capsule repertoire.
Let's get the code review out of the way, but I don't want to merge it until the experience with BrainFrame is better.