# VGGPerceptualLoss in mixed precision mode

**URL:** <https://discuss.pytorch.org/t/vggperceptualloss-in-mixed-precision-mode/88704>\
**Category:** vision\
**Created:** [July 10, 2020, 3:32pm UTC](https://discuss.pytorch.org/t/vggperceptualloss-in-mixed-precision-mode/88704 "2020-07-10T15:32:59Z")\
**Posts on this page:** 1\
**Showing post:** 6

<div class="post-metadata">

**Author:** ![FilipAndersson245](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/filipandersson245/32/14909_2.png) [@FilipAndersson245](https://discuss.pytorch.org/u/FilipAndersson245)\
**Post date:** [August 7, 2020, 3:11pm UTC](https://discuss.pytorch.org/t/vggperceptualloss-in-mixed-precision-mode/88704/6 "2020-08-07T15:11:00Z")

</div>

I fixed the oversaturation by changing how i did the inv\_norm. The fix me and @ZimoNitrome did can be seen in [here](https://discuss.pytorch.org/t/how-to-normalize-a-tensor-to-0-mean-and-1-variance/18766/19), mixed precision still produces noise and fail to converge.

---

_[View the full topic](https://discuss.pytorch.org/t/vggperceptualloss-in-mixed-precision-mode/88704)._