How to config the object detection pipeline config to train a ssd-mobilenet model using the ssd512 architecture instead of ssd300
Answers
Answered by
1
Answer:
The object detection config should be make using ssd-mobilnet model.
Explanation:
The highest [email protected] score is 0.68, considering the difficulty of the SPID dataset, the score is not bad.
The model fails to detect, in most missing detect cases, relatively small objects when error detection was run.
The lack of capability to detect the small objects is the most common problem of the process of deploying SSD object detection.
However the ssd512 model is much more helpful than ssd300.
In the pipeline config the image_resizer section direct from 300 to 512. The base architecture of the ssd512 is not same as the ssd300.
Similar questions