def transform_image(image):
# TODO: Hardcoded ImageNet dataset mean
image = np.array(image) - np.array([103.939, 116.779, 123.68])
image = image / np.array([57.375, 57.12, 58.395])
image = image.transpose((2, 0, 1))
image = image[np.newaxis, :]
return image
def highlightBackProp(image, model2):
#steering_angle = float(model.predict(image[None, :, :, :], batch_size=1))
# print(image.shape)
#image = cv2.resize(image, (256, 256) )
oldimage = image
#image = cv2.resize(image, (200, 66) )
# transpose if model is other way
count, h, w, ch = model2.inputs[0].get_shape()
ih, iw, ich = image.shape
if h == ich and ch == ih:
image = image.transpose()
#print(image.shape)
m1d = model2.predict(image[None, :, :, :], batch_size=1)
#print(m1d.shape)
m1d = np.squeeze(m1d, axis=0)
m1d = np.squeeze(m1d, axis=2)
#m1d = cv2.resize(image, (120, 160) )
#m1d=np.resize(m1d,(120,160))
#print(m1d.shape)
#print(m1d)
#plt.hist(m1d[::-1])
#plt.show()
#print(m1d.max())
#print(m1d.min())
o2 = overlay = Image.fromarray(cm.Reds(m1d / m1d.max(), bytes=True))
#o2= o2.convert("RGB")
#return o2
#plt.imshow(o2);
#plt.show();
pixeldata = list(overlay.getdata())
for i, pixel in enumerate(pixeldata):
if almostEquals(pixel[:3], (255, 255, 255)):
pixeldata[i] = (255, 255, 255, 0)
else:
pixeldata[i] = (pixel[0], pixel[1], pixel[2], 128)
overlay.putdata(pixeldata)
#obig= cv2.resize(overlay, (320, 160) )
carimg = Image.fromarray(np.uint8(image))
#carimg = Image.fromarray(np.uint8(oldimage))
carimg = carimg.convert("RGBA")
new_img2 = Image.alpha_composite(carimg, overlay)
new_img2 = new_img2.convert("RGB")
o2 = o2.convert("RGB")
#plt.imshow(o2);
#plt.show();
return np.array(new_img2)
def _evaluate_model(self, generator, model, iou_threshold, score_threshold,
max_detections, save_path):
"""
Evaluate a given dataset using a given model.
Parameters
----------
generator : The generator that represents the dataset to evaluate.
model : The model to evaluate.
iou_threshold : The threshold used to consider when a detection is positive or negative.
score_threshold : The score confidence threshold to use for detections.
max_detections : The maximum number of detections to use per image.
save_path : The path to save images with visualized detections to.
Returns
-------
all_detections : A list containing the predicted boxes for each image in the generator.
"""
box_list = []
score_list = []
for i in range(generator.size()):
raw_image = generator.load_image(i)
image = generator.preprocess_image(raw_image.copy())
image, scale = generator.resize_image(image)
if keras.backend.image_data_format() == 'channels_first':
image = image.transpose((2, 0, 1))
# run network
boxes, scores, labels = model.predict_on_batch(
np.expand_dims(image, axis=0))[:3]
# correct boxes for image scale
boxes /= scale
for box, score in zip(boxes[0], scores[0]):
if score < 0.5:
break
b = box.astype(int)
box_list.append(b)
score_list.append(score)
### !!! SAVEPATH CURRENTLY NOT IMPLEMENTED !!!
### This optional feature can be added later, maybe for TA3, allowing images to be output with
### bounding boxes to a specified directory after evaluation.
# if save_path is True:
# draw_annotations(raw_image, generator.load_annotations(i), label_to_name = generator.label_to_name)
# draw_detections(raw_image, image_boxes, image_scores, image_labels, label_to_name = generator.label_to_name, score_threshold = score_threshold)
# cv2.imwrite(os.path.join(save_path, '{}.png'.format(i)), raw_image)
return box_list, score_list
def on_post(self, req, res):
"""Handles POST requests"""
data = req.get_param('file').file.read()
# アプロードされたファイルを保存する
# f = request.files['file']
#filepath = "./static/" + datetime.now().strftime("%Y%m%d%H%M%S") + ".jpg"
#data.save(filepath)
pilimg = Image.open(io.BytesIO(data))
# x = np.asarray(pilimg)
# モデルを使って判定する
### load model and weight
model = model_from_json(open('apple_orange_model.json').read())
model.load_weights('apple_orange_weights.h5')
# 画像を読み込み、グレースケールに変換し、28x28pixelに変換し、numpy配列へ変換する。
# 画像の1ピクセルは、それぞれが0-255の数値。
# image = np.array(Image.open(filepath).convert("L").resize((28, 28)))
image = np.array(pilimg.resize((25, 25)))
image = image.transpose(2, 0, 1)
image = image.reshape(1, image.shape[0] * image.shape[1] *
image.shape[2]).astype("float32")[0]
x = np.array([image / 255.])
result = model.predict_classes(x)
proba = model.predict_proba(x)
# image = np.array(pilimg.convert("L").resize((28, 28)))
# print(filepath)
# さらにフラットな1次元配列に変換。
# image = image.reshape(1, 784).astype("float32")[0]
# result = model.predict_classes(np.array([image / 255.]))
# print("result:", result[0], "（0:りんご, 1:オレンジ）")
predict = result[0].tolist()
result = classes[predict]
predict_proba = proba[0].tolist()
result_proba = predict_proba[predict]
# クラスを予測
# 入力は1枚の画像なので[0]のみ
# pred = model.predict(x)[0]
# 予測確率が高いトップ5を出力
#top = 5
#top_indices = pred.argsort()[-top:][::-1]
#result = classes[top_indices[0]]
# 予測確率が高いトップ1を出力
res.status = falcon.HTTP_200
res.body = json.dumps({'result': result, 'probability': result_proba})
def rotate_images(image):
'''
rotate_images(image):
This function rotates an image on it's center 7 times (45, 90, 135, 180, 225, 270, and mirror image)
Input:
image: One image file
Returns:
A list of images containing the original image and the rotated one
'''
rotated_images = []
chirl_image = image.transpose(Image.FLIP_LEFT_RIGHT)
rotated_images.extend([image, chirl_image])
return rotated_images
def upload_file():
if request.method == 'GET':
return render_template('index.html')
if request.method == 'POST':
# アプロードされたファイルを保存する
#f = request.files['file']
#filepath = "./static/" + datetime.now().strftime("%Y%m%d%H%M%S") + ".jpg"
#f.save(filepath)
# モデルを使って判定する
# data = req.get_param('file').file.read()
data = request.files['file'].read()
pilimg = Image.open(io.BytesIO(data))
### load model and weight
model = model_from_json(open('apple_orange_model.json').read())
model.load_weights('apple_orange_weights.h5')
image = np.array(pilimg.resize((25, 25)))
image = image.transpose(2, 0, 1)
image = image.reshape(1, image.shape[0] * image.shape[1] * image.shape[2]).astype("float32")[0]
x = np.array([image / 255.])
result = model.predict_classes(x)
proba = model.predict_proba(x)
predict = result[0].tolist()
result = classes[predict]
predict_proba = proba[0].tolist()
result_proba = predict_proba[predict]
response = jsonify({'result':result, 'probability':result_proba})
# res.status = falcon.HTTP_200
response.status_code = 200
return response
def MetaFunct(i):
df = readJson(i)
df = flatten_json(df)
a = pd.DataFrame.from_dict(df, orient='index')
a.columns = ['Values']
label = a[a.index.str.match('label')]
if label.shape[0]>0:
label.index = label.index.str.split('_',expand=True)
label.reset_index(inplace=True)
label.drop(columns=['level_0'], inplace=True)
label = label[label.level_2.str.match('score')]
label.sort_values(by=['Values'], ascending=[0], inplace=True)
label.reset_index(inplace=True)
label.drop(columns=['index'], inplace=True)
if label.shape[0] > 4:
label = label.iloc[:5,:]
label['level_1'] = label['level_2'] + '_' + label['level_1'].astype(str)
label.drop(columns=['level_2'], inplace = True)
label = pd.melt(label, id_vars=['level_1'])
label.drop(columns=['variable'], inplace = True)
label.set_index('level_1', inplace=True)
label = label.transpose()
label.reset_index(inplace=True, drop=True)
image = a[a.index.str.match('image')]
if image.shape[0]>0:
image.index = image.index.str.split('_',expand=True)
image.reset_index(inplace=True)
image.reset_index(inplace=True)
image.drop(columns= ['index','level_0','level_1','level_2'], inplace=True)
image['level_4'] = image['level_4'] + '_' + image['level_5'].astype(str)
image.drop(columns=['level_5'], inplace=True)
image = pd.pivot_table(image, index='level_3', columns='level_4', values='Values',aggfunc='first' ).reset_index()
if image.shape[0] > 4:
image = image.iloc[:5,:]
image = pd.melt(image, id_vars=['level_3'])
image['level_3'] = image['level_4'] + '_' + image['level_3'].astype(str)
image.drop(columns=['level_4'], inplace=True)
image.set_index('level_3', inplace=True)
image = image.transpose()
image.reset_index(inplace=True, drop=True)
crop = a[a.index.str.match('crop')]
if crop.shape[0]>0:
crop.index = crop.index.str.split('_',expand=True)
crop.reset_index(inplace=True)
crop.drop(columns=['level_0','level_1', 'level_2'], inplace=True)
crop['level_3'] = crop['level_3'] + '_' + crop['level_6'].astype(str)
crop.drop(columns=['level_4'], inplace=True)
crop.drop(columns=['level_6'], inplace=True)
bounding = crop[crop.level_3.str.match('bounding')]
if bounding.shape[0] > 0:
bounding['level_3'] = bounding['level_3'] + '_' + bounding['level_5']
bounding.drop(columns=['level_5'], inplace=True)
bounding.set_index('level_3', inplace=True)
bounding = bounding.transpose()
bounding.reset_index(inplace=True, drop=True)
confidence = crop[crop.level_3.str.match('confidence')]
if crop.shape[0]>0:
confidence = pd.DataFrame(confidence)
confidence = confidence['Values']
confidence = pd.DataFrame(confidence)
confidence.columns = np.array(['Confidence'])
confidence.reset_index(inplace=True, drop=True)
importance = crop[crop.level_3.str.match('importance')]
if importance.shape[0]>0 :
importance = crop[crop.level_3.str.match('importance')]
importance = importance['Values']
importance.reset_index(inplace=True, drop=True)
importance = pd.DataFrame(importance)
importance.columns = np.array(['Importance'])
crophints = pd.concat([bounding,confidence,importance], axis=1)
print(i)
document= pd.DataFrame({"PetID":[(i.split('.')[0]).split('/')[2]]})
final = pd.concat([document, label, image, crophints], axis=1)
return(final)
def _evaluate_model(self, generator, model, iou_threshold, score_threshold, max_detections, save_path):
"""
Evaluate a given dataset using a given model.
Parameters
----------
generator : The generator that represents the dataset to evaluate.
model : The model to evaluate.
iou_threshold : The threshold used to consider when a detection is positive or negative.
score_threshold : The score confidence threshold to use for detections.
max_detections : The maximum number of detections to use per image.
save_path : The path to save images with visualized detections to.
Returns
-------
all_detections : A list containing the predicted boxes for each image in the generator.
"""
#all_detections = [[None for i in range(generator.num_classes()) if generator.has_label(i)] for j in range(generator.size())]
#print(np.shape(all_detections), file = sys.__stdout__)
box_list = []
score_list = []
for i in range(0,3):
raw_image = generator.load_image(i)
image = generator.preprocess_image(raw_image.copy())
image, scale = generator.resize_image(image)
if keras.backend.image_data_format() == 'channels_first':
image = image.transpose((2, 0, 1))
# run network
boxes, scores, labels = model.predict_on_batch(np.expand_dims(image, axis = 0))[:3]
# correct boxes for image scale
boxes /= scale
for box, score in zip(boxes[0], scores[0]):
if score < 0.5:
break
b = box.astype(int)
box_list.append(b)
score_list.append(score)
# select indices which have a score above the threshold
# indices = np.where(scores[0, :] > score_threshold)[0]
# # select those scores
# scores = scores[0][indices]
# # find the order with which to sort the scores
# scores_sort = np.argsort(-scores)[:max_detections]
# # select detections
# image_boxes = boxes[0, indices[scores_sort], :]
# image_scores = scores[scores_sort]
# image_labels = labels[0, indices[scores_sort]]
# image_detections = np.concatenate([image_boxes, np.expand_dims(image_scores, axis = 1), np.expand_dims(image_labels, axis = 1)], axis = 1)
# # if save_path is True:
# # draw_annotations(raw_image, generator.load_annotations(i), label_to_name = generator.label_to_name)
# # draw_detections(raw_image, image_boxes, image_scores, image_labels, label_to_name = generator.label_to_name, score_threshold = score_threshold)
# # cv2.imwrite(os.path.join(save_path, '{}.png'.format(i)), raw_image)
# # copy detections to all_detections
# for label in range(generator.num_classes()):
# if not generator.has_label(label):
# continue
# all_detections[i][label] = image_detections[image_detections[:, -1] == label, :-1]
#print(score_list, file = sys.__stdout__)
return box_list, score_list