Update classifier_classify_new.py

src/embedding/classifier_classify_new.py

@@ -42,6 +42,9 @@
 #import judgeutil
 
 BASEDIR = os.getenv('RUNTIME_BASEDIR',os.path.abspath(os.path.dirname(__file__)))
+
+
+HAS_OPENCL = os.getenv('HAS_OPENCL','true')
 sys.path.append(BASEDIR)
 import judgeutil
 
@@ -363,7 +366,10 @@ def train_svm_with_embedding(args_list):
         judge_paths = paths
         judge_labels = labels
     judge_nrof_images = len(judge_paths)
-    judge_emb_array = np.zeros((judge_nrof_images, 512))
+    if HAS_OPENCL == 'true:
+        judge_emb_array = np.zeros((judge_nrof_images, 512))
+    else:
+        judge_emb_array = np.zeros((judge_nrof_images, 128))
     for j in range(judge_nrof_images):
         judge_embedding = None
         image_path = judge_paths[j]
@@ -390,7 +396,10 @@ def train_svm_with_embedding(args_list):
     nrof_images = len(paths)
     nrof_batches_per_epoch = int(math.ceil(1.0 * nrof_images / args.batch_size))
 
-    emb_array = np.zeros((nrof_images, 512))
+    if HAS_OPENCL == 'true':
+        emb_array = np.zeros((nrof_images, 512))
+    else:
+        emb_array = np.zeros((nrof_images, 128))
     for i in range(nrof_batches_per_epoch):
         start_index = i*args.batch_size
         end_index = min((i+1)*args.batch_size, nrof_images)