Update how_sr3_works.ipynb

Author: sdash77

guide/14-deep-learning/how_sr3_works.ipynb

@@ -172,7 +172,11 @@
     "* `linear_start` - Optional integer. schedule start. default set to `1e-05`.\n",
     "* `linear_end` - Optional integer. schedule end. default set to `1e-02`.\n",
     "\n",
-    "Default values are set for 64 -> 256 upscale tasks. Number of res blocks can be decreased when 'out-of-memory' issue is encountered. Optimal results are generally obtained, if trained for atleast 200-300 epochs provided there is adequate amount of data.\n",
+    "#### recommendations:\n",
+    "- Default values are set for 64 -> 256 upscale tasks. for landsat/sentinel or coarser resolution tasks, values from 64 -> 512 task gives better results.\n",
+    "- Number of res blocks can be decreased if 'out-of-memory' issue is encountered. \n",
+    "- Optimal results are generally obtained, if trained for atleast 200-300 epochs provided there is adequate amount of data.\n",
+    "- Its observed that larger norm_groups often cause hue deviation, decreasing often fixes the issue. \n",
     "\n",
     "Here, `data` is the object returned from `prepare_data` function. \n",
     "\n",