Coding Period: Weeks 4 and 5
Overview
Week 4 had been a VERYY BADD week for me, for personal reasons (basically overburdening myself with different stupid tasks, and not managing to get out anything in any task) and this hampered my work REALLY BAD. I wasn’t able to work at all during this week, and my mentors proposed a kinda ultimatum: We’ll give you 2 days extension and you’ll produce results, as the Midterm evualations are approaching quickly, and I didn’t have much to show for it. It was decided that the fourth week SHOULD be the last week for the finding of a model. However, due to my sheer incompetence, this was dragged in my fifth week as well.
Objectives
- Check the codebase as well
- Close the ssd_mobilenetv2 model
- Show the running model
- [] Use seaborn to get prettier plots
- [] Updating the graph in real time
The execution
The first thing that I decided was to close the ssd_mobilenetv2,but decided to use the ssdlite_mobilenetv2 model, as it’s apparently just faster that the non-lite model, and then, i went to find a model trained on the coco dataset (no specific reason, just knew that the COCO dataset also covers humans in it’s object detection process). And after finding it on the tensorflow object detection site [7], i moved on to convert it to the ONNX format. Since I had already done the process for the SSD_volumenetv1, i decided to just reuse that, since both the versions contain the same input and output shapes and dimensions. So after obtaining the ONNX model, i decided to run the RoboticsAcademy exercise with it, I found out that the this model was getting a better benchmarking score than the previous one (36.2% mAP vs. 35% mAP), which was to be expected, since it’s just an upgrade (a higher version) to the previous model. However, the other 3 parts of the exercise didn’t work on my model, which I’m pretty sure is a problem in my laptop, a not in the exercise itself. Next, i went through the codebase as well, by following the Readme [8], as it explained how RoboticsAcademy was built and the interactions between all the different parts work.
References
[1] https://github.com/JdeRobot/RoboticsAcademy
[2] https://onnxruntime.ai/
[3] https://www.khronos.org/nnef
[4] https://github.com/onnx/models
[5] https://github.com/KhronosGroup/NNEF-Tools/tree/master/models#nnef-model-zoo
[6] https://www.tensorflow.org/tutorials\
[7] https://github.com/tensorflow/models/tree/master/research/object_detection/models
[8] https://github.com/JdeRobot/RoboticsAcademy/blob/master/docs/InstructionsForDevelopers.md\