example training done

denoisplit/notebooks/example_training.ipynb

@@ -5,13 +5,15 @@
    "metadata": {},
    "source": [
     "# Objective\n",
-    "In this notebook, we tackle the problem of joint denoising and splitting, which has a clear usecase with fluorescence microscopy. For this, we first train [denoiSplit](https://arxiv.org/pdf/2403.11854.pdf) and later evaluate it. The overall schema is shown below:\n",
+    "In this notebook, we tackle the problem of joint splitting and unsupervised denoising, which has a usecase with fluorescence microscopy. From a technical perspective, given a noisy image $x$, the goal is to predict two images $c_1$ and $c_2$ such that $x = c_1 + c_2$. In other words, we have a superimposed image $x$ and we want to predict the denoised estimates of the constituent images $c_1$ and $c_2$. It is important to note that the network is trained with noisy data and the denoising is done in a unsupervised manner. \n",
+    "\n",
+    "For this, we will use [denoiSplit](https://arxiv.org/pdf/2403.11854.pdf), a recently developed approach for this task. In this notebook we train denoiSplit and later evaluate it on one validation frame. The overall schema for denoiSplit is shown below:\n",
     "<!-- Insert a figure -->\n",
     "<!-- ![Schema](teaser.png) -->\n",
     "<img src=\"teaser.png\" alt=\"drawing\" width=\"800\"/>\n",
     "\n",
     "\n",
-    "For this purpose, we look at CCPs vs ER task, one of the tasks tackled by denoiSplit which is generated from [BioSR](https://figshare.com/articles/dataset/BioSR/13264793).\n",
+    "Here, we look at CCPs (clathrin-coated pits) vs ER (Endoplasmic reticulum) task, one of the tasks tackled by denoiSplit which is generated from [BioSR](https://figshare.com/articles/dataset/BioSR/13264793) dataset. For this task, the noise is synthetically added. \n",
     "\n",
     "What to change: There are a couple of hyper-parameters present in the config creation section. You are encouraged to read the description of each hyper-parameter and change them to see how it affects the performance of the model. Note that one needs to restart and run all cells after changing the hyper-parameters."
    ]