Still vastly underused in image to image to this day and I don't understand why: Unets and self-attention. I keep seeing papers come out that have obvious problems that could be solved using these two simple things!
-
I'm curious to see what U-Nets with EfficientNet backbone can do. I was also reading the DeepLabv3 paper, it seems that its dilated convolutions are wonderful to preserve fine grained details. It looks like U-Nets win doing blob-like segmentation while DeepLabv3 wins on details.
2/ I'm not sure exactly what's going on but it seems like the EfficientNets are just a bit lacking in the kinds of feature detection that image to image thrives on. What I wind up seeing is more zombie limbs, for example.
-
-
Oh that is quite interesting, I wouldn’t have thought that ResNet would be better than EfficientNet, especially B3 or B4
-
Yeah me neither. To be clear- we actually made a working mobile model with B4 so it does work. It's just quite disappointing honestly relative to what you'd think you'd get out of it based on benchmarks.
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.