Assigned
Status Update
No update yet.
Comments
ma...@gmail.com
I have forwarded this request to the engineering team. We will update this issue with any progress updates and a resolution.
Best Regards,
Josh Moyer
Google Cloud Platform Support
Best Regards,
Josh Moyer
Google Cloud Platform Support
nr...@google.com
nr...@google.com
This is not only useful for IP addresses, but also for many other resources. I understand that names are currently used as identifiers, so this request is probably not trivial to implement. Maybe distinguishing between a (numeric, automatically generated) identifier and a (textual) label is the way to go?
Description
E.g., I have tested the image of a black widow that I have attached here (testpic.jpg)
With attached sample image, Vision label detection returns following overly generic labels:
- Insect 93%
- Arthropod 92%
- Spider 86%
- Terrestrial Animal 80%
- Pest 77%
Now, compare this, to the top 5 highest confidence results of a classifier based on EfficientNetV2XL (Trained on ImageNet Full 2011 release 21841 classes):
- black_widow, Latrodectus_mactans
- arachnid, arachnoid
- invertebrate
- comb-footed_spider, theridiid
- arthropod
As you can see these are perfectly accurate contrarily to Vision results that are too generic, not even mentioning that the highest confidence result is "insect" that is not only generic but also wrong considering that spiders aren't insects.
I expect more from the paid service of a company like Google that has contributed a lot to computer vision field.