Fix to allow Scheduler to be configured as a list and fix a bug in the AdaptiveTrainingCallback

[harimkang]

Summary

• Fix wrong update the frequency of the warmup-scheduler (This is the cause of a bug from AdaptiveTrainingCallback. -> Longer warmup times in small-dataset)

Model	Dataset	Before (map_50)	After (map_50)	seed
ATSS_MobilenetV2	Pothole_small_0	0.0	0.46	0

• Fixes an issue where overrides would throw an error if configs.callbacks came in as None.
• Modify the optimizer and scheduler to receive them as a list.
• WarmupReduceLROnPlateau -> LinearWarmupScheduler + ReduceLROnPlateau.
• WarmupPolynomialLR -> LinearWarmupScheduler + PolynomialLR.
• It also modifies it to match the configuration based on OTX1.6 (develop).
  • max_epochs, warmup_iters, early-stopping patience

How to test

Checklist

• I have added unit tests to cover my changes.​
• I have added integration tests to cover my changes.​
• I have added e2e tests for validation.
• I have added the description of my changes into CHANGELOG in my target branch (e.g., CHANGELOG in develop).​
• I have updated the documentation in my target branch accordingly (e.g., documentation in develop).
• I have linked related issues.

License

• I submit my code changes under the same Apache License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2023 Intel Corporation
# SPDX-License-Identifier: Apache-2.0

[sovrasov]

@harimkang could you also add results on tests/assets/car_tree_bug/ to increase confidence that the fix can handle various datasets?

[harimkang]

@harimkang could you also add results on tests/assets/car_tree_bug/ to increase confidence that the fix can handle various datasets?

For now, it seems to be working fine based on atss_mobilenetv2. I've also run a regression test and will share the results when I get them.

[eugene123tw]

@harimkang Could you also check the accuracy for instance seg algos on car_tree_bug dataset?

[harimkang]

@harimkang Could you also check the accuracy for instance seg algos on car_tree_bug dataset?

@eugene123tw
With car_tree_bug dataset

	test/map_50	best-epochs
maskrcnn_efficientnetb2b	1	19
maskrcnn_r50	1	14

I'll also attach the regression result when it's done.

[harimkang]

[Small-Dataset Comparison]
WW06.3: e3cc66f (3 times)
This PR: 53f5dcb (1 time)
classification (multi-class): multiclass-CUB-small
detection: pothole_small
*WW06.3 is the average of three measurements, so it may vary slightly.

Task	model	ww	test/score	train/iter_time	duration
MULTI_CLASS_CLS	Efficientnet_b0	ww06.3	0.9529		8
		This PR	0.9492		6.8
	Efficientnet_v2	ww06.3	0.9962		13
		This PR	0.9831		15.6333
	MobileNet_v3_large_1.0	ww06.3	0.9737		7
		This PR	0.9887		6.8333
	DeiT-Tiny	ww06.3	0.4915		8
		This PR	1		7.7
	DINO-v2	ww06.3	0.9661		11
		This PR	0.9661		13.4667
DETECTION	yolox_x	ww06.3	0.2834	0.1651	56
		This PR	0.3015	0.1558	44
	yolox_l	ww06.3	0.0069	0.1763	37
		This PR	0.3613	0.174	45
	yolox_s	ww06.3	0	0.1181	18
		This PR	0.1594	0.1049	35
	yolox_tiny	ww06.3	0.0022	0.0948	19
		This PR	0.3862	0.0917	24
	ssd_mobilenetv2	ww06.3	0.006	0.1175	20
		This PR	0.2115	0.1137	23
	atss_resnext101	ww06.3	0.494	0.2825	75
		This PR	0.5272	0.2831	70
	atss_mobilenetv2	ww06.3	0	0.1548	18
		This PR	0.534	0.171	31

[sungmanc]

Thanks for the fix