Model for UFO Object Types based on Shapes (Team 6)

models/ufo_models/README.md

@@ -0,0 +1,86 @@
+# Steps:
+The folder contains the graph(.pb), the output_labels.txt and the checkpoint files
+Using Shape as the classifying parameter for retraining objects we have utilised the following **12 shapes**:
+
+**Blimp, cigar, circle, cylinder, disc, fireball, oval, saturn-like, sphere, square, star-like, triangle.**
+
+We extracted unique images per shape and then out of the bunch 12 shape files **having more than 30 images** (which is required for the internal partition for train,test and verification stages)
+
+**Steps for the retraining of the Inception V3 model for the objects:**
+
+1   Install tensorflow Hub
+````
+$ pip install "tensorflow>=1.7.0"
+$ pip install tensorflow-hub
+````
+2   Activate tensorflow
+````
+source ~/tensorflow/bin/activate # bash, sh, ksh, or zsh
+````
+3   Generate the input folder containing the sub-folders used for retraining
+ufostalkerimages (the images were resized wrt aspect ratio)
+
+4   Run the retrainer script:
+(https://github.com/tensorflow/hub/raw/r0.1/examples/image_retraining/retrain.py)
+````
+python retrain.py --image_dir ~/path_to_ufostalkerimages
+````
+5   This generates the retrained model that can be visualized on the Tensorboard.
+
+**Steps to deploy the retrained model using the label_image script:**
+
+(https://github.com/tensorflow/tensorflow/raw/master/tensorflow/examples/label_image/label_image.py)
+
+````
+python label_image.py \
+--graph=/tmp/output_graph.pb 				//graph that contains the retrained model
+--labels=/tmp/output_labels.txt \			//output labels are the shapes like blimp,cigar,circle etc.
+--input_layer=Placeholder \
+--output_layer=final_result \
+--image=path_to_image_to_be_classified
+````
+ **Following is a sample input and output snippet:**
+````
+(tensorflow) sayali@sayali-VirtualBox:~/Desktop/example_code$ python label_image.py \
+ --graph=/tmp/output_graph.pb --labels=/tmp/output_labels.txt \
+ --input_layer=Placeholder \
+ --output_layer=final_result \
+ --image=$HOME/Desktop/ufostalkerimages/Blimp/82805.jpg
+2018-04-12 02:33:51.419390: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
+blimp 0.93306917
+saturn like 0.012200303
+sphere 0.010724315
+fireball 0.009860759
+triangle 0.00829932
+````
+
+
+**Integration with the current docker file:**
+The tika-dockers repo contains the InceptionRestDockerFile that holds the current training model namely Inception V4.
+The Path to inception_v4.ckpt & meta files needs to be replaced with the custom paths to our retrained model.ckpt and meta files.
+
+**Performance**
+The resultant model has good performance with respect to the accuracy of detecting the UFO objects specifically emphasizing and utilizing the shape characteristic of the UFO sightings from the aggregated dataset.
+
+
+**Testing results on a generic image from Google(input: Triangle-shaped UFO ):**
+Source URL:.dailymail.co.uk/i/pix/2017/12/04/14/46FA2D2B00000578-0-image-a-103_1512399265747.jpg
+
+````
+(tensorflow) sayali@sayali-VirtualBox:~/Desktop/example_code$ python label_image.py --graph=/tmp/output_graph.pb --labels=/tmp/output_labels.txt --input_layer=Placeholder --output_layer=final_result --image=$HOME/Desktop/tri.jpg
+2018-04-13 02:19:12.347016: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
+triangle 0.24518733
+fireball 0.20667171
+oval 0.16336635
+star like 0.13473332
+sphere 0.07973641
+````
+**Future Scope**
+
+1  The second model for Captions can be retrained by replacing the nouns with the objects(retrained with shapes) in the existing sentence corpus to generate more meaningful captions relevant to the UFO Object sightings.
+
+2  Models can be retrained with the MobileNet_V2 architecture for higher speeds and deployment on mobile and IOT devices.
+
+
+
+

models/ufo_models/output_labels.txt

@@ -0,0 +1,12 @@
+blimp
+cigar
+circle
+cylinder
+disc
+fireball
+oval
+saturn like
+sphere
+square
+star like
+triangle