fix: input size setting

[jkbmrz]

Adding a fix to the parsers that require model input size and cannot deduce it from the model outputs (YuNetParser and LaneDetectionParser) by extending the build() method to look for the the input_shape within the head_config. Additionally, ParserGenerator is updated to set the input_shape argument aromatically based on the relevant NN Archive.

Some additional changes:

• the default input_size was removed for the LaneDetectionParser,
• all input_shape variable/parameter names were changed to input_size because it better fits their meaning (shape resembles NCHW/NHWC whereas we only describe HW) → this is more or less a refactoring change only, but it also effects the input_size setters for the two parsers (setInputShape → setInputSize)

The implemented changes were tested on YuNet and Ultra Fast Lane Detection models.

[github-actions]

☂️ Python Coverage

current status: ✅

Overall Coverage

Lines	Covered	Coverage	Threshold	Status
3748	1249	33%	0%	🟢

New Files

No new covered files...

Modified Files

File	Coverage	Status
depthai_nodes/ml/parsers/base_parser.py	70%	🟢
depthai_nodes/ml/parsers/lane_detection.py	20%	🟢
depthai_nodes/ml/parsers/utils/yunet.py	15%	🟢
depthai_nodes/ml/parsers/yunet.py	15%	🟢
depthai_nodes/parser_generator.py	24%	🟢
TOTAL	29%	🟢

updated for commit: 1ed14ba by action🐍

[github-actions]

Test Results

125 tests   125 ✅  1s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit 1ed14ba.

♻️ This comment has been updated with latest results.

[klemen1999]

Generally this method could work but I wonder if there isn't a nicer way to infer input size based on the shape of the output and maybe some known stride information or something similar? We did it with YOLO but might not be applicable for all models so if this is not the case here then you can disregard this comment.
If we go ahead with using input_size let's make sure we update this in the whole package accordingly.

[jkbmrz]

Generally this method could work but I wonder if there isn't a nicer way to infer input size based on the shape of the output and maybe some known stride information or something similar?

I found no way of determining input size for YuNetParser and LaneDetection models (the problem is that we don't know the size ratio of the inputs and thus cannot reconstruct it based on the flattened outputs).

If we go ahead with using input_size let's make sure we update this in the whole package accordingly.

Let's wait for others to see what's their position on it and I can then proceed with changing it!

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 2.63158% with 37 lines in your changes missing coverage. Please review.

Please upload report for BASE (main@94a3043). Learn more about missing BASE report.

✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
depthai_nodes/ml/parsers/lane_detection.py	5.26%	18 Missing ⚠️
depthai_nodes/ml/parsers/yunet.py	0.00%	10 Missing ⚠️
depthai_nodes/parser_generator.py	0.00%	5 Missing ⚠️
depthai_nodes/ml/parsers/utils/yunet.py	0.00%	4 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #126   +/-   ##
=======================================
  Coverage        ?   33.27%           
=======================================
  Files           ?       68           
  Lines           ?     3754           
  Branches        ?        0           
=======================================
  Hits            ?     1249           
  Misses          ?     2505           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.