[Object_Detection] Add BoxCoder for SSD and FasterRCNN #4

tanzhenyu · 2020-06-16T22:52:55Z

Add BoxCoder for SSD and FasterRCNN.

bhack · 2020-06-17T13:36:21Z

Just as reminder we have a boxcoder in official/vision

tanzhenyu · 2020-06-17T15:30:04Z

Just as reminder we have a boxcoder in official/vision

We do, this is to "graduate" that API, but as a Keras object

bhack · 2020-06-17T16:29:30Z

We do, this is to "graduate" that API, but as a Keras object

Ok I suppose that this "graduation" could be not API compatible with official/vision so how do you handle this in the spirit of the approved RFC already mentioned at tensorflow/community#223 (comment)?
Will be in voluntary/optional charge to the model garden maintainers to refactor after graduation? Will be part of the graduation process itself?
If not, will this target just t+1 models (so no official models refactoring)? Or will be reserved/mandatory only to models that will be contributed to keras-cv?

This just to understand the level of duplicates minimization that we are aspiring and future merging policy for models by an user/contributor point of view.

tanzhenyu · 2020-06-17T17:27:36Z

We do, this is to "graduate" that API, but as a Keras object

Ok I suppose that this "graduation" could be not API compatible with official/vision so how do you handle this in the spirit of the approved RFC already mentioned at tensorflow/community#223 (comment)?
Will be in voluntary/optional charge to the model garden maintainers to refactor after graduation? Will be part of the graduation process itself?
If not, will this target just t+1 models (so no official models refactoring)? Or will be reserved/mandatory only to models that will be contributed to keras-cv?

This just to understand the level of duplicates minimization that we are aspiring and future merging policy for models by an user/contributor point of view.

So this is exactly being mentioned in:
https://github.com/tensorflow/community/blame/master/rfcs/20190802-model-garden-redesign.md#L60-L61

The tentative proposal is to have them in keras-CV, not addons, and can be contributed by model-garden team as well, so we wouldn't be having duplicating low-level APIs such as this in vision/. The proposal is that they will directly make a dependency on this, which is lightweight enough.

bhack · 2020-06-17T17:38:54Z

The tentative proposal is to have them in keras-CV, not addons, and can be contributed by model-garden team as well, so we wouldn't be having duplicating low-level APIs such as this in vision/. The proposal is that they will directly make a dependency on this, which is lightweight enough.

Yes but you have not replied on who is in charge of this process? In a similar case it seems to me more clear tensorflow/community#260 (comment)

tanzhenyu · 2020-06-17T18:06:12Z

The tentative proposal is to have them in keras-CV, not addons, and can be contributed by model-garden team as well, so we wouldn't be having duplicating low-level APIs such as this in vision/. The proposal is that they will directly make a dependency on this, which is lightweight enough.

Yes but you have not replied on who is in charge of this process? In a similar case it seems to me more clear tensorflow/community#260 (comment)

We will probably release the finalized plan soon, which should answer your question. Stay tuned :-)

saberkun · 2020-06-19T15:54:00Z

kerascv/layers/ssd_box_coder.py

+        $ \hat{width_gt} = log(width_gt / width_a)
+        $ \hat{height_gt} = log(height_gt / height_a)
+
+    where cx, cy, width, height represents center of width, center of height, width, height respectively,


Do you have a coding style setting up? It looks this line is too long.

We should follow usual conventions and keep line length below 85 char. Note that black will not format comments/docstrings.

All that's why :-) Done

saberkun · 2020-06-19T15:54:33Z

tests/kerascv/layers/ssd_box_coder_test.py

+from kerascv.layers.ssd_box_coder import SSDBoxCoder
+
+
+def test_encode_decode_variance():


Do we want to test all modes (e.g. graph, eager) of keras?

Is that testing util exposed?

Do you mean like in https://github.com/tensorflow/addons/blob/master/tensorflow_addons/utils/test_utils.py?

fchollet

Thank you for the PR. Let's get a review from the model garden team as well so we can make sure we have a path forward for how this block will be used in the model garden codebase.

fchollet · 2020-06-19T17:24:47Z

kerascv/layers/ssd_box_coder.py

+        $ \hat{width_gt} = log(width_gt / width_a)
+        $ \hat{height_gt} = log(height_gt / height_a)
+
+    where cx, cy, width, height represents center of width, center of height, width, height respectively,


We should follow usual conventions and keep line length below 85 char. Note that black will not format comments/docstrings.

fchollet · 2020-06-19T17:25:32Z

kerascv/layers/ssd_box_coder.py

+class SSDBoxCoder(tf.keras.layers.Layer):
+    """Defines a SSDBoxEncoder that converts encodes the ground_truth_boxes using anchors.
+
+    Mathematically, the encoding is:


In general, don't use LaTeX in docstrings, as it is hard to read before rendering (and for many people even after rendering). Prefer using Python-like pseudocode.

fchollet · 2020-06-19T20:50:07Z

kerascv/layers/ssd_box_coder.py

+    `ground_truth_boxes` to anchors based on a certain matching strategy (argmax, bipartite)
+
+    # Attributes:
+        variances: The 1-D scaling factor with 4 floats. This is used to represent the variance of


It may be preferable to have 4 keyword arguments instead (or potentially 2 tuple arguments, one for the centers and one for the height/width), as it would avoid confusion regarding the order of the 4 floats

fchollet

LGTM for API

…d update iou losses (#1296) * first attempt at introducing YoloX * formatted and fixed bugs * cast fix #1 * cast fix #2 * cast fix #3 * cast fix #4 * adding ensure shape for support * reverting and removing ensure_shape * fixed another bug * updated train.py * updated docs, tests and added support for loss strings * first attempt at introducing YoloX * formatted and fixed bugs * adding ensure shape for support * reverting and removing ensure_shape * reformatted by black * fixed a linting issue * finally rebased atop the recent changes * finally rebased atop the new changes * fixed linting issues * reverted rebasing issues with iou loss * fixing rebased errors part 2 * fixed more linting issues * TPU testing changes * linting fixes * updated with implementation details from paper * updated based on review comments and api changes * first attempt at introducing YoloX * updated docs, tests and added support for loss strings * fixed linting issues * reverted rebasing issues with iou loss * review comments * removed examples * linting fix * fixed rebasing error * updated no_reduction warning * review comments * revert version and linting fixes

Port CSPDarkNet and DarkNet to Keras Core

…d update iou losses (keras-team#1296) * first attempt at introducing YoloX * formatted and fixed bugs * cast fix keras-team#1 * cast fix keras-team#2 * cast fix keras-team#3 * cast fix keras-team#4 * adding ensure shape for support * reverting and removing ensure_shape * fixed another bug * updated train.py * updated docs, tests and added support for loss strings * first attempt at introducing YoloX * formatted and fixed bugs * adding ensure shape for support * reverting and removing ensure_shape * reformatted by black * fixed a linting issue * finally rebased atop the recent changes * finally rebased atop the new changes * fixed linting issues * reverted rebasing issues with iou loss * fixing rebased errors part 2 * fixed more linting issues * TPU testing changes * linting fixes * updated with implementation details from paper * updated based on review comments and api changes * first attempt at introducing YoloX * updated docs, tests and added support for loss strings * fixed linting issues * reverted rebasing issues with iou loss * review comments * removed examples * linting fix * fixed rebasing error * updated no_reduction warning * review comments * revert version and linting fixes

tanzhenyu added 2 commits June 16, 2020 15:48

Add BoxEncoder for SSD and FasterRCNN.

cb663df

[Object_Detection] Rename BoxEncoder to BoxCoder.

37a9993

tanzhenyu requested a review from fchollet June 16, 2020 22:52

Additional comment in docstring.

c7bdc8f

saberkun reviewed Jun 19, 2020

View reviewed changes

fchollet reviewed Jun 19, 2020

View reviewed changes

[Object_Detection] Address comments for SSDBoxCoder.

fb7463d

fchollet approved these changes Jun 22, 2020

View reviewed changes

tanzhenyu closed this Sep 12, 2020

tanzhenyu deleted the box_coder branch September 17, 2020 22:43

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Oct 30, 2022

cast fix keras-team#4

ff4f62a

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Nov 3, 2022

cast fix keras-team#4

7041015

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Jan 20, 2023

cast fix keras-team#4

51d9512

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Jan 29, 2023

cast fix keras-team#4

d22df9b

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Feb 8, 2023

cast fix keras-team#4

8c68957

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Mar 22, 2023

cast fix keras-team#4

7fbd1f5

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Mar 22, 2023

cast fix keras-team#4

cf269d4

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Mar 27, 2023

cast fix keras-team#4

e92ac03

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Mar 29, 2023

cast fix keras-team#4

a24f39c

quantumalaviya added a commit to quantumalaviya/keras-cv that referenced this pull request Apr 24, 2023

cast fix keras-team#4

a4aef19

tirthasheshpatel pushed a commit to tirthasheshpatel/keras-cv that referenced this pull request Jun 5, 2023

Merge pull request keras-team#4 from ianstenbit/core-csp

0fcd6a8

Port CSPDarkNet and DarkNet to Keras Core

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Object_Detection] Add BoxCoder for SSD and FasterRCNN #4

[Object_Detection] Add BoxCoder for SSD and FasterRCNN #4

tanzhenyu commented Jun 16, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

saberkun Jun 19, 2020

fchollet Jun 19, 2020

tanzhenyu Jun 19, 2020

saberkun Jun 19, 2020 •

edited

Loading

tanzhenyu Jun 19, 2020

bhack Jun 19, 2020

fchollet left a comment

fchollet Jun 19, 2020

fchollet Jun 19, 2020

tanzhenyu Jun 19, 2020

fchollet Jun 19, 2020

tanzhenyu Jun 19, 2020

fchollet left a comment

		from kerascv.layers.ssd_box_coder import SSDBoxCoder


		def test_encode_decode_variance():

[Object_Detection] Add BoxCoder for SSD and FasterRCNN #4

[Object_Detection] Add BoxCoder for SSD and FasterRCNN #4

Conversation

tanzhenyu commented Jun 16, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

bhack commented Jun 17, 2020

tanzhenyu commented Jun 17, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

saberkun Jun 19, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

fchollet left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

fchollet left a comment

Choose a reason for hiding this comment

saberkun Jun 19, 2020 •

edited

Loading