from predictor import ScoringService as ss
if __name__ == '__main__':
print('Model Load:', ss.get_model())
print(ss.predict(ss.get_inputs(), '2020-12-30'))
Do not include other categories in the prediction you will make.
- If you do not want to make any prediction in some frames,
just write "prediction = {}" in the prediction of the frame in the sequence(in line 65 or line 67).
"""
#評価用の動画データをNo.0〜順番に読み込み
test_data_path = '../data'
videos = sorted(glob.glob(test_data_path + '/*.mp4'))
# 検出したBounding Boxの座標を出力する
# Output = [name_label, BoxA_X_min, BoxA_X_max, BoxA_Y_min, BoxA_Y_max]
Output_list = ''
for i in range(len(videos)):
video_path = videos[i]
ScoringService.get_model()
Output = ScoringService.predict(video_path)
print(Output)
if i == 0: #最初はキーを指定して辞書作成
Output_list = Output
else: #2個目以降はキーを指定して辞書追加
Output_list.update(Output)
print("＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊")
# Output douga mp4
#ScoringService.get_model()
#ScoringService.pw_outdouga(video_path)
with open('../output/prediction.json', 'w') as f:
# The video file on which we want to perform inference (object detection and tracking) is passed as file path parameter to the predict function. # The results are returned in json format consisting of class predictions and assigned ids for each frame. # Multiple videos are passed by looping over the predict function and the results are aggregated into one json file. # Released under Apache License 2.0 # Email: [email protected] # ----------------------------------------------------------- import os os.environ["CUDA_VISIBLE_DEVICES"] = "0" import pdb import json import time from predictor import ScoringService exp_name = "flip_lr.aspect." if ScoringService.get_model(): start_time = time.time() target_videos = [ "train_00", "train_01", "train_02", "train_12", "train_16", "train_22" ] # target_videos = ["train_{:02d}".format(i) for i in range(0,25)] combined_prediction_json = {} print("Processing videos = {}".format(target_videos)) for train_file in target_videos: start_time_train = time.time() print("Train_file = {}".format(train_file)) preds_json = ScoringService.predict( "/ext/signate_edge_ai/train_videos/{}.mp4".format(train_file)) with open(exp_name + '{}.preds.json'.format(train_file), 'w+') as output_json_file:
from predictor import ScoringService as cls # データセットをダウンロードして解凍したファイルを配置した場所を定義します。 # データ保存先ディレクトリ DATASET_DIR= "../../data_dir" # 読み込むファイルを定義します。 inputs = cls.get_inputs(DATASET_DIR) print(inputs) cls.train_and_save_model(inputs, model_path="model")
- The categories for testing are "Car" and "Pedestrian".
Do not include other categories in the prediction you will make.
- If you do not want to make any prediction in some frames,
just write "prediction = {}" in the prediction of the frame in the sequence(in line 65 or line 67).
"""
data_path = '../data'#読みこむデータのパスを記載
#複数のファイルに対応済み
videos = glob.glob(data_path+'/*.mp4')
Output_list = ''
for i in range(len(videos)):
video_path = videos[i]
ScoringService.get_model()
#Output = ScoringService.predict(video_path)
#print(Output)
#if i == 0:#最初はキーを指定して辞書作成
# Output_list = Output
#else:#2個目以降はキーを指定して辞書追加
# Output_list.update(Output)
print("＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊＊")
# Output douga mp4
#ScoringService.get_model()
ScoringService.pw_outdouga(video_path)
#with open('../output/prediction.json', 'w') as f: