docs: add CellVIT finetuning notebook example

Author: okunator

changelog.d/20230826_115035_oskari.lehtonen.md

@@ -0,0 +1,2 @@
+## Docs
+- Add notebook example on finetuning CellVIT-SAM with accelerate.

examples/pannuke_nuclei_segmentation_cellvit.ipynb

@@ -215,7 +215,7 @@
     "All of these are defined in the `train()`-wrapper function.\n",
     "\n",
     "**The model**.\n",
-    "First, we will define the CellPose nuclei segmentation model with a `imagenet` pre-trained encoder. Specifically, we will use the `focalnet_small_lrf` backbone for this demonstration. Many more encoders can be used, since these are imported from the `timm` library. There are also support for some transformer based encoders, but these are shown in another notebooks.\n",
+    "First, we will define the OmniPose nuclei segmentation model with a `imagenet` pre-trained encoder. Specifically, we will use the `focalnet_small_lrf` backbone for this demonstration. Many more encoders can be used, since these are imported from the `timm` library. There are also support for some transformer based encoders, but these are shown in another notebooks.\n",
     "\n",
     "**Branch losses**.\n",
     "For each output of the model, we define a joint-loss function. These losses are summed together during backprop to form a multi-task loss. For the `\"omnipose\"`branch output we set MAE-loss (Mean Absolute Error) and for the `\"type\"` (cell type predictions) outputs we will set a joint-loss composed of CE-loss (Cross Entropy) and DICE-loss for both. For the CE-losses, we will also be using [spectral decoupling](https://arxiv.org/abs/2011.09468) to regularize the model. \n",