概述:
代码获取:可私信
在TensorFlow中实现单镜头多盒检测器(SSD),用干检测和分类交通标志。该实现能够在具有 Intel Corei7-6700K的GTX1080实现40-45fos
请注意,此项目仍在进行中。现在的主要问题是模型过度拟合,
我目前正在先进行VOC2012的预培训,然后进行交通标志检测的转移学习。目前只检测到停车标志和人行横道标志。检测图像示例如下。
依赖库与代码
Skip to content Product Solutions Open Source Pricing Search Sign in Sign up georgesung / ssd_tensorflow_traffic_sign_detection Public Code Issues 32 Pull requests Actions Projects Security Insights ssd_tensorflow_traffic_sign_detection/inference.py / @georgesung georgesung Removed unused function run_inference_old() Latest commit 88f1781 on Feb 15, 2017 History 1 contributor 189 lines (155 sloc) 6.08 KB ''' Run inference using trained model ''' import tensorflow as tf from settings import * from model import SSDModel from model import ModelHelper from model import nms import numpy as np from sklearn.model_selection import train_test_split import cv2 import math import os import time import pickle from PIL import Image import matplotlib.pyplot as plt from moviepy.editor import VideoFileClip from optparse import OptionParser import glob def run_inference(image, model, sess, mode, sign_map): """ Run inference on a given image Arguments: * image: Numpy array representing a single RGB image * model: Dict of tensor references returned by SSDModel() * sess: TensorFlow session reference * mode: String of either "image", "video", or "demo" Returns: * Numpy array representing annotated image """ # Save original image in memory image = np.array(image) image_orig = np.copy(image) # Get relevant tensors x = model['x'] is_training = model['is_training'] preds_conf = model['preds_conf'] preds_loc = model['preds_loc'] probs = model['probs'] # Convert image to PIL Image, resize it, convert to grayscale (if necessary), convert back to numpy array image = Image.fromarray(image) orig_w, orig_h = image.size if NUM_CHANNELS == 1: image = image.convert('L') # 8-bit grayscale image = image.resize((IMG_W, IMG_H), Image.LANCZOS) # high-quality downsampling filter image = np.asarray(image) images = np.array([image]) # create a "batch" of 1 image if NUM_CHANNELS == 1: images = np.expand_dims(images, axis=-1) # need extra dimension of size 1 for grayscale # Perform object detection t0 = time.time() # keep track of duration of object detection + NMS preds_conf_val, preds_loc_val, probs_val = sess.run([preds_conf, preds_loc, probs], feed_dict={x: images, is_training: False}) if mode != 'video': print('Inference took %.1f ms (%.2f fps)' % ((time.time() - t0)*1000, 1/(time.time() - t0))) # Gather class predictions and confidence values y_pred_conf = preds_conf_val[0] # batch size of 1, so just take [0] y_pred_conf = y_pred_conf.astype('float32') prob = probs_val[0] # Gather localization predictions y_pred_loc = preds_loc_val[0] # Perform NMS boxes = nms(y_pred_conf, y_pred_loc, prob) if mode != 'video': print('Inference + NMS took %.1f ms (%.2f fps)' % ((time.time() - t0)*1000, 1/(time.time() - t0))) # Rescale boxes' coordinates back to original image's dimensions # Recall boxes = [[x1, y1, x2, y2, cls, cls_prob], [...], ...] scale = np.array([orig_w/IMG_W, orig_h/IMG_H, orig_w/IMG_W, orig_h/IMG_H]) if len(boxes) > 0: boxes[:, :4] = boxes[:, :4] * scale # Draw and annotate boxes over original image, and return annotated image image = image_orig for box in boxes: # Get box parameters box_coords = [int(round(x)) for x in box[:4]] cls = int(box[4]) cls_prob = box[5] # Annotate image image = cv2.rectangle(image, tuple(box_coords[:2]), tuple(box_coords[2:]), (0,255,0)) label_str = '%s %.2f' % (sign_map[cls], cls_prob) image = cv2.putText(image, label_str, (box_coords[0], box_coords[1]), 0, 0.5, (0,255,0), 1, cv2.LINE_AA) return image def generate_output(input_files, mode): """ Generate annotated images, videos, or sample images, based on mode """ # First, load mapping from integer class ID to sign name string sign_map = {} with open('signnames.csv', 'r') as f: for line in f: line = line[:-1] # strip newline at the end sign_id, sign_name = line.split(',') sign_map[int(sign_id)] = sign_name sign_map[0] = 'background' # class ID 0 reserved for background class # Create output directory 'inference_out/' if needed if mode == 'image' or mode == 'video': if not os.path.isdir('./inference_out'): try: os.mkdir('./inference_out') except FileExistsError: print('Error: Cannot mkdir ./inference_out') return # Launch the graph with tf.Graph().as_default(), tf.Session() as sess: # "Instantiate" neural network, get relevant tensors model = SSDModel() # Load trained model saver = tf.train.Saver() print('Restoring previously trained model at %s' % MODEL_SAVE_PATH) saver.restore(sess, MODEL_SAVE_PATH) if mode == 'image': for image_file in input_files: print('Running inference on %s' % image_file) image_orig = np.asarray(Image.open(image_file)) image = run_inference(image_orig, model, sess, mode, sign_map) head, tail = os.path.split(image_file) plt.imsave('./inference_out/%s' % tail, image) print('Output saved in inference_out/') elif mode == 'video': for video_file in input_files: print('Running inference on %s' % video_file) video = VideoFileClip(video_file) video = video.fl_image(lambda x: run_inference(x, model, sess, mode, sign_map)) head, tail = os.path.split(video_file) video.write_videofile('./inference_out/%s' % tail, audio=False) print('Output saved in inference_out/') elif mode == 'demo': print('Demo mode: Running inference on images in sample_images/') image_files = os.listdir('sample_images/') for image_file in image_files: print('Running inference on sample_images/%s' % image_file) image_orig = np.asarray(Image.open('sample_images/' + image_file)) image = run_inference(image_orig, model, sess, mode, sign_map) plt.imshow(image) plt.show() else: raise ValueError('Invalid mode: %s' % mode) if __name__ == '__main__': # Configure command line options parser = OptionParser() parser.add_option('-i', '--input_dir', dest='input_dir', help='Directory of input videos/images (ignored for "demo" mode). Will run inference on all videos/images in that dir') parser.add_option('-m', '--mode', dest='mode', default='image', help='Operating mode, could be "image", "video", or "demo"; "demo" mode displays annotated images from sample_images/') # Get and parse command line options options, args = parser.parse_args() input_dir = options.input_dir mode = options.mode if mode != 'video' and mode != 'image' and mode != 'demo': assert ValueError('Invalid mode: %s' % mode) if mode != 'demo': input_files = glob.glob(input_dir + '/*.*') else: input_files = [] generate_output(input_files, mode)
Python 3.5+
TensorFlow v0.12.0
Pickle
OpenCV Python
Matplotlib(可选)
运用
将此存储库克降到某处,让我们将其称为$ROOT
从头开始训练模型:
代码流程
※Download the LISA Traffic Sign Dataset, and store it in a directory $LISA_DATA ※cd $LISA_DATA ※Follow instructions in the LISA Traffic Sign Dataset to create 'mergedAnnotations.csv' such that only stop signs and pedestrian ※crossing signs are shown ※cp $ROOT/data_gathering/create_pickle.py $LISA_DATA ※python create_pickle.py ※cd $ROOT ※ln -s $LISA_DATA/resized_images_* . ※ln -s $LISA_DATA/data_raw_*.p . ※python data_prep.py ※This performs box matching between ground-truth boxes and default ※boxes, and packages the data into a format used later in the ※pipeline ※python train.py ※This trains the SSD model ※python inference.py -m demo
效果
如上所述,该SSD实现能够在具有Intel Core i7 6700K的GTX 1080上实现40-45 fps
推理时间是神经网络推理时间和非最大抑制(NMS)时间的总和。总的来说,神经网络推断时间明显小干NMS时间,神经网络推理时间通常在7-8ms之间,而NMS时间在15-16ms之间。这里实现的NMS算法尚未优化,仅在CPU上运行,因此可以在那里进一步努力提高性能。
数据集
整个LISA交通标志数据集由47个不同的交通标志类别组成。因为我们只关注这些类的子集,所以我们只使用LSA数据集的子集。此外,我们忽略了没有找到匹配的默认框的所有训练样本,从而进一步减小了数据集的大小。由于这个过程,我们最终只能处理很少的数据。
为了改进这一问题,我们可以执行图像数据增强,和/或在更大的数据集上预训练模型(例如 VOC2012、ILSVRC)
下载链接:】(https://cvrr.ucsd.edu/LISA/lisa-traffic-sign-dataset.html)
代码可私信
代码可私信
代码可私信