Reviews for Wingman Jr. Filter
Wingman Jr. Filter by Zephyr
Response by Zephyr
Developer response
posted 5 years agoYou're most definitely welcome!
Regarding that bug: that is, as far as I know, a limitation of the underlying webRequest API. I do not believe that there is a way to hook that when the URL is visited directly via that API, and I haven't looked into other API's to try to do so. I hadn't set up a dedicated repository with issues for just the plugin yet, and you gave me a great reason to do so! I'm now tracking your bug/request over here: https://github.com/wingman-jr-addon/wingman_jr/issues/1
(UPDATE: I'm wrong! You can do this with the "main_frame" webRequestType. Watch issue for more details.)
(UPDATE 2: Issue has a proposed fix, slated for next release!)
(UPDATE 3: Fixed in latest release!)
Regarding the model: I'm actually not using nsfwjs for this. It's currently a MobileNetV2 finetune on a dataset built from scratch. The underlying model itself is on four categories: safe, questionable, racy, and explicit; the plugin blends those into the number you see presented. If you're interested in further discussion on exposing model internals etc. please post a GitHub issue and we can continue there.
Thanks for taking the time to review the addon and give some feedback!
Regarding that bug: that is, as far as I know, a limitation of the underlying webRequest API. I do not believe that there is a way to hook that when the URL is visited directly via that API, and I haven't looked into other API's to try to do so. I hadn't set up a dedicated repository with issues for just the plugin yet, and you gave me a great reason to do so! I'm now tracking your bug/request over here: https://github.com/wingman-jr-addon/wingman_jr/issues/1
(UPDATE: I'm wrong! You can do this with the "main_frame" webRequestType. Watch issue for more details.)
(UPDATE 2: Issue has a proposed fix, slated for next release!)
(UPDATE 3: Fixed in latest release!)
Regarding the model: I'm actually not using nsfwjs for this. It's currently a MobileNetV2 finetune on a dataset built from scratch. The underlying model itself is on four categories: safe, questionable, racy, and explicit; the plugin blends those into the number you see presented. If you're interested in further discussion on exposing model internals etc. please post a GitHub issue and we can continue there.
Thanks for taking the time to review the addon and give some feedback!