Merge branch 'master' of https://github.com/leoarc/tfjs-models

README.md

@@ -12,7 +12,9 @@ The following sample classification models can be used by caMicroscope's predict
 
 ## Segmentation Models
 The following segmentation models can be used by caMicroscope's segmentation algorithm, as an alternative to traditional CV methods. Segmentation models take in image data and return a mask of the same size. When using the model with caMicroscope, the whether the model operates on black and white or RGB images should be specified.
-* [Mitosis](egmentation%20Sample%20Models/mitosis-segmentation) -- [model](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/model.json), [weights part 1](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/group1-shard1of2.bin), and [weights part 2](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/group1-shard2of2.bin). 
+* [Mitosis](Segmentation%20Sample%20Models/mitosis-segmentation) -- [model](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/model.json), [weights part 1](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/group1-shard1of2.bin), and [weights part 2](Segmentation%20Sample%20Models/mitosis-segmentation/tfjs_model/group1-shard2of2.bin). 
+
+* [YNet](Segmentation%20Sample%20Models/YNet) -- [model](Segmentation%20Sample%20Models/YNet/tfjs_model/model.json), [weights part 1](Segmentation%20Sample%20Models/YNet/tfjs_model/group1-shard1of2.bin), and [weights part 2](Segmentation%20Sample%20Models/YNet/tfjs_model/group1-shard2of2.bin). 
 
 ## Common Issues:
 - ***Provided weight data has no target variable***: This might be a common issue for the model to fail to load. To avoid this, make sure that the keras model is loaded exactly once. To make sure, reset the runtime and load the model before carrying to conversion. This case also goes when your converting using the bash command. The Keras model which you saved must be loaded just once. To make sure, save the model, reset the runtime, load the model and save it again. Multiple loads seem to mess up the layer names. You can track this issue further [here](https://github.com/tensorflow/tfjs/issues/755).

Segmentation Sample Models/YNet/README.md

@@ -0,0 +1,33 @@
+# YNet
+## Dataset 
+The dataset used is available in [this](https://www.bcsc-research.org/data/variables/) post.
+
+A total of 87 pathologists diagnosed a randomly assigned subset of 60 slides into four diagnostic categories producing an average of 22 diagnostic labels per case.  The average size of these ROIs is 10,000×12,000. Out of these 200 ROIs have been used.
+
+The four diagnostic categories are : i) benign ii) atypia  iii) ductal carcinomain situ iv) invasive cancer
+
+Patch Size used : 256 X 256
+
+Image Format : RGB
+
+Pre-Processing : 256 X 256 patches are cropped from the high redolution ( 10000 X 12000 ) images without any overlap. As there is no seperate test dataset 15% of the extracted patches are kept aside for testing and the rest are used for training. Pixel values are normalized before training.
+
+Magnification : 100x
+
+## Model
+[Y-net](https://arxiv.org/pdf/1806.01313.pdf): 
+
+![model architecture](ynet.png)
+
+
+
+Citation 
+
+``` @inproceedings{mehta2018ynet,
+  title={{Y-Net: Joint Segmentation and Classification for Diagnosis of Breast Biopsy Images}},
+  author={Sachin Mehta and Ezgi Mercan and Jamen Bartlett and Donald Weaver and Joann  Elmore and Linda Shapiro},
+  booktitle={International Conference on Medical image computing and computer-assisted intervention},
+  year={2018},
+  organization={Springer}
+}
+```

Segmentation Sample Models/mitosis-segmentation/README.md

@@ -10,4 +10,4 @@ Pixel values are normalized between 0-1.
 ## Model
 U-net: 256 -> 128 -> 64 -> 32 -> 16 -> 32 -> 64 -> 128 -> 256.
 
-Input image size: 256 x 256 x 3
+Input image size: 256 x 256 x 3