Preparing for model release

PyTorchClassification/classify_image.py

@@ -0,0 +1,61 @@
+################
+#
+# classify_image.py
+#
+# Test driver for running inference with the ClassificationModel class in models.py.
+#
+################
+
+#%% Imports
+
+from models import ClassificationModel
+import argparse
+import sys
+
+
+#%% Functions
+
+def run_classification_model(model_path,image_path):
+    
+    # [560, 560] for ensemble, [560] or [448] for single models like ResNeXt
+    model_input_sizes = [560,560]
+
+    model = ClassificationModel(model_path, model_input_sizes, useGPU=True)
+    species, vals = model.predict_image(image_path, 3)
+
+    for i in range(0, len(species)):
+        print('%d) %s\tlikelihood: %f' % (i+1, species[i], vals[i]))
+    
+
+#% Command-line driver
+    
+def main():
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument('model_path', type=str, help='Model filename')
+    parser.add_argument('image_path', type=str, help='Image filename')
+    
+    if len(sys.argv[1:])==0:
+        parser.print_help()
+        parser.exit()
+        
+    args = parser.parse_args()    
+    run_classification_model(args.model_path,args.image_path)
+
+    
+if __name__ == '__main__':
+    
+    main()
+
+
+#%% Interactive driveer
+    
+if False:
+
+    #%%    
+    
+    model_path = r"D:\temp\species_classification\sc_all_extended_ensemble_resnext_inceptionV4_560_2019.08.27_model.pytorch"
+    image_path = r"D:\temp\species_classification\190215-meer-full.jpg"
+    run_classification_model(model_path,image_path)
+    
+    

PyTorchClassification/models.py

@@ -11,10 +11,6 @@
 ################
 
 
-import os
-import shutil
-import time
-import copy
 from enum import Enum
 import numpy as np
 import pretrainedmodels

PyTorchClassification/onnx_test_driver.py

@@ -1,17 +1,28 @@
-#%% Constants and imports
-
-# Note to self:
+###
+#
+# onnx_test_driver.py
+#
+# Example driver to run an image through the .onnx version of our species 
+# classification model.
+#
+# Installation notes:
+#
 # conda install pytorch-nightly -c pytorch
 # pip install future
 # pip install opencv-python
 # pip install onnx
-
+# pip install caffe2
+#
 # Useful links:
+#
 # https://caffe2.ai/docs/tutorial-loading-pre-trained-models.html
 # https://github.com/onnx/tutorials/blob/master/tutorials/OnnxCaffe2Import.ipynb
+#
+###
+
+#%% Constants and imports
 
 import cv2
-from caffe2.python import core, workspace
 import onnx
 import caffe2.python.onnx.backend
 import numpy as np
@@ -20,15 +31,15 @@ from operator import itemgetter
 
 TOP_K = 5
 
-IMAGE_FILENAME = '/data/images/lion.jpg'
-MODEL_FILENAME = '/data/models/exported_model.onnx'
-CLASSLIST_FILENAME = '/data/models/classlist.txt'
+IMAGE_FILENAME = r"D:\temp\species_classification\190215-meer-full.jpg"
+MODEL_FILENAME = r"D:\temp\species_classification\sc_all_extended_ensemble_resnext_inceptionV4_560_2019.08.27_model.onnx"
+CLASSLIST_FILENAME = r"D:\temp\species_classification\sc_all_extended_ensemble_resnext_inceptionV4_560_2019.08.27_classes.txt"
 
-# Target mean / std; should match the values used at training time
 MODEL_IMAGE_SIZE = 224
 MODEL_RESIZE_SIZE = 256
 OVERSIZE_FACTOR = 1.3
 
+
 #%% Load and prepare image (using cv2)
 
 # Load image
@@ -65,15 +76,19 @@ imgExpanded = np.expand_dims(imgFinal,0).astype('float32')
 
 # cv2.imshow('image',imgCropped); cv2.waitKey(0); cv2.destroyAllWindows()
 
+
 #%% Load model
+
 model = onnx.load(MODEL_FILENAME)
 
+
 #%% Run model
 
 # Run the ONNX model with Caffe2
 outputs = caffe2.python.onnx.backend.run_model(model, [imgExpanded])
 outputs_softmax = outputs[0][0].astype(np.float64)
 
+
 #%% Print top K class names
 
 # Load class names

PyTorchClassification/test.py

@@ -1,25 +0,0 @@
-################
-#
-# test.py
-#
-# Test driver for running inference with the ClassificationModel class in models.py.
-#
-################
-
-from models import *
-
-def main():
-
-    image_path = '/path/to/image.jpg'
-    model_path = '/path/to/model.pth.tar'
-    # [560, 560] for ensemble, [560] or [448] for single models like ResNeXt
-    model_input_sizes = [560]
-
-    model = ClassificationModel(model_path, model_input_sizes, useGPU = True)
-    species, vals = model.predict_image(image_path, 3)
-
-    for i in range(0, len(species)):
-        print('%d) %s\tlikelihood: %f' % (i+1, species[i], vals[i]))
-
-if __name__ == '__main__':
-    main()

classify_images.py

@@ -15,9 +15,6 @@
 # * a taxonomy file, so the scientific names used in the training data can
 #   be mapped to common names.
 #
-# We are currently testing against PyTorch 0.4.1 and Cuda 9.0, and we have tested on 
-# both Linux and Windows.
-#
 ####### 
 
 
@@ -35,15 +32,14 @@ API_ROOT = r'd:\git\SpeciesClassification'
 # Path to taxa.csv, for latin --> common mapping
 #
 # Set to None to disable latin --> common mapping
-TAXONOMY_PATH = r'd:\temp\taxa.csv' # None
+TAXONOMY_PATH = r'd:\temp\species_classification\taxa.19.08.28.0536.csv' # None
 
 IMAGES_TO_CLASSIFY = [
-        r"D:\temp\animals\African_Elephant\30651.ngsversion.1421960098780.jpg",
-        r"D:\temp\animals\Alligator\Alligator_mississippiensis_01.JPG"
+        r"D:\temp\species_classification\190215-meer-full.jpg",
+        r"D:\temp\species_classification\African_Bush_Elephant.jpg"
         ]
 
-# CLASSIFICATION_MODEL_PATH = r'd:\temp\models\inc4-incres2-560-78.5\model_deploy.pth.tar'
-CLASSIFICATION_MODEL_PATH = r"D:\temp\models\resnext-448-78.8\model_best.pth.tar"
+CLASSIFICATION_MODEL_PATH = "D:\temp\species_classification\sc_all_extended_ensemble_resnext_inceptionV4_560_2019.08.27_model.pytorch"
 
 # Detection (i.e., bounding box generation) is optional; set to None 
 # to disable detection
@@ -61,8 +57,8 @@ USE_GPU = True
 # We typically specify [560,560] if we're loading our Inception/InceptionResnet 
 # ensemble. For ResNext, we typically specify [448].
 #
-# IMAGE_SIZES = [560, 560]
-IMAGE_SIZES = [448]
+IMAGE_SIZES = [560, 560]
+# IMAGE_SIZES = [448]
 
 
 #%% Path setup to import the classification code