#!/usr/bin/python3 print("Preparing...") import tensorflow as tf import os from tqdm import tqdm import generate import forward import cv2 import numpy as np import mvsdk print("Finish!") def save_kernal(fp, val): print(val.shape[2], file=fp) print(val.shape[3], file=fp) print(val.shape[1], file=fp) print(val.shape[0], file=fp) for in_channel in range(val.shape[2]): for out_channel in range(val.shape[3]): for row in range(val.shape[0]): for col in range(val.shape[1]): print(val[row][col][in_channel][out_channel], file=fp) def save_weight_mat(fp, val): print(val.shape[0], file=fp) print(val.shape[1], file=fp) for row in range(val.shape[0]): for col in range(val.shape[1]): print(val[row][col], file=fp) def save_bias(fp, val): print(val.shape[0], file=fp) for i in range(val.shape[0]): print(val[i], file=fp) def save_para(folder, paras, names, info): os.system("mkdir %s/%s" % (folder, info)) for para, name in zip(paras, names): fp = open("%s/%s/%s" % (folder, info, name), "w") if name[-1:] == "b": save_bias(fp, para) elif name[:2] == "fc": save_weight_mat(fp, para) elif name[:4] == "conv": save_kernal(fp, para) fp.close() STEPS = 100000 BATCH = 40 LEARNING_RATE_BASE = 0.0003 LEARNING_RATE_DECAY = 0.99 MOVING_AVERAGE_DECAY = 0.99 def train(dataset, show_bar=False): x = tf.placeholder(tf.float32, [None, generate.SRC_ROWS, generate.SRC_COLS, generate.SRC_CHANNELS]) y_ = tf.placeholder(tf.float32, [None, forward.OUTPUT_NODES]) keep_rate = tf.placeholder(tf.float32) nodes, vars, vars_name = forward.forward(x, 0.01, keep_rate) y = nodes[-1] ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1)) # ce = tf.nn.weighted_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1), pos_weight=1) cem = tf.reduce_mean(ce) loss = cem + tf.add_n(tf.get_collection("losses")) global_step = tf.Variable(0, trainable=False) learning_rate = tf.train.exponential_decay( LEARNING_RATE_BASE, global_step, len(dataset.train_samples) / BATCH, LEARNING_RATE_DECAY, staircase=False) train_step = tf.train.AdamOptimizer(learning_rate).minimize(loss, global_step=global_step) ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step) ema_op = ema.apply(tf.trainable_variables()) with tf.control_dependencies([train_step, ema_op]): train_op = tf.no_op(name='train') correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True)) with tf.Session(config=config) as sess: init_op = tf.global_variables_initializer() sess.run(init_op) bar = tqdm(range(STEPS), ascii=True, dynamic_ncols=True) for i in bar: images_samples, labels_samples = dataset.sample_train_sets(BATCH, 0.03) _, loss_value, step = sess.run( [train_op, loss, global_step], feed_dict={x: images_samples, y_: labels_samples, keep_rate: 0.3} ) if step % 500 == 0: test_images, test_labels = dataset.sample_test_sets(10000) test_acc, output = sess.run([accuracy, y], feed_dict={x: test_images, y_: test_labels, keep_rate: 1.0}) output = np.argmax(output, axis=1) real = np.argmax(test_labels, axis=1) print("=============test-set===============") for n in range(forward.OUTPUT_NODES): print("label: %d, precise: %f, recall: %f" % (n, np.mean(real[output == n] == n), np.mean(output[real == n] == n))) train_images, train_labels = dataset.sample_train_sets(10000) train_acc, output = sess.run([accuracy, y], feed_dict={x: train_images, y_: train_labels, keep_rate: 1.0}) output = np.argmax(output, axis=1) real = np.argmax(train_labels, axis=1) print("=============train-set===============") for n in range(forward.OUTPUT_NODES): print("label: %d, precise: %f, recall: %f" % (n, np.mean(real[output == n] == n), np.mean(output[real == n] == n))) print("\n") if train_acc >= 0.99 and test_acc >= 0.99: vars_val = sess.run(vars) save_para( "model", vars_val, vars_name, "steps:%d-train_acc:%f-test_acc:%f" % (step, train_acc, test_acc) ) bar.set_postfix({"loss": loss_value, "train_acc": train_acc, "test_acc": test_acc}) # print("save done!") # pred = sess.run(y, feed_dict={x: test_images, keep_rate:1.0}) # nodes_val = sess.run(nodes, feed_dict={x:test_images}) # return vars_val, nodes_val DevList = mvsdk.CameraEnumerateDevice() nDev = len(DevList) if nDev < 1: print("No camera was found!") return for i, DevInfo in enumerate(DevList): print("{}: {} {}".format(i, DevInfo.GetFriendlyName(), DevInfo.GetPortType())) i = 0 if nDev == 1 else int(input("Select camera: ")) DevInfo = DevList[i] print(DevInfo) # 打开相机 hCamera = 0 try: hCamera = mvsdk.CameraInit(DevInfo, -1, -1) except mvsdk.CameraException as e: print("CameraInit Failed({}): {}".format(e.error_code, e.message)) return # 获取相机特性描述 cap = mvsdk.CameraGetCapability(hCamera) # 判断是黑白相机还是彩色相机 monoCamera = (cap.sIspCapacity.bMonoSensor != 0) # 黑白相机让ISP直接输出MONO数据，而不是扩展成R=G=B的24位灰度 if monoCamera: mvsdk.CameraSetIspOutFormat(hCamera, mvsdk.CAMERA_MEDIA_TYPE_MONO8) else: mvsdk.CameraSetIspOutFormat(hCamera, mvsdk.CAMERA_MEDIA_TYPE_BGR8) # 相机模式切换成连续采集 mvsdk.CameraSetTriggerMode(hCamera, 0) # 手动曝光，曝光时间30ms mvsdk.CameraSetAeState(hCamera, 0) mvsdk.CameraSetExposureTime(hCamera, 30 * 1000) # 让SDK内部取图线程开始工作 mvsdk.CameraPlay(hCamera) # 计算RGB buffer所需的大小，这里直接按照相机的最大分辨率来分配 FrameBufferSize = cap.sResolutionRange.iWidthMax * cap.sResolutionRange.iHeightMax * (1 if monoCamera else 3) # 分配RGB buffer，用来存放ISP输出的图像 # 备注：从相机传输到PC端的是RAW数据，在PC端通过软件ISP转为RGB数据（如果是黑白相机就不需要转换格式，但是ISP还有其它处理，所以也需要分配这个buffer） pFrameBuffer = mvsdk.CameraAlignMalloc(FrameBufferSize, 16) while (cv2.waitKey(1) & 0xFF) != ord('q'): # 从相机取一帧图片 try: pRawData, FrameHead = mvsdk.CameraGetImageBuffer(hCamera, 200) mvsdk.CameraImageProcess(hCamera, pRawData, pFrameBuffer, FrameHead) mvsdk.CameraReleaseImageBuffer(hCamera, pRawData) # 此时图片已经存储在pFrameBuffer中，对于彩色相机pFrameBuffer=RGB数据，黑白相机pFrameBuffer=8位灰度数据 # 把pFrameBuffer转换成opencv的图像格式以进行后续算法处理 frame_data = (mvsdk.c_ubyte * FrameHead.uBytes).from_address(pFrameBuffer) frame = np.frombuffer(frame_data, dtype=np.uint8) frame = frame.reshape((FrameHead.iHeight, FrameHead.iWidth, 1 if FrameHead.uiMediaType == mvsdk.CAMERA_MEDIA_TYPE_MONO8 else 3)) frame = cv2.resize(frame, (640, 480), interpolation=cv2.INTER_LINEAR) cv2.imshow("Press q to end", frame) if (cv2.waitKey(1) & 0xFF) == ord(' '): roi = cv2.selectROI("roi", frame) roi = frame[roi[1]:roi[1] + roi[3], roi[0]:roi[0] + roi[2]] print(roi) cv2.imshow("box", roi) image = cv2.resize(roi, (48, 36)) image = image.astype(np.float32) / 255.0 out = sess.run(y, feed_dict={x: [image]}) print(out) print(np.argmax(out)) except mvsdk.CameraException as e: if e.error_code != mvsdk.CAMERA_STATUS_TIME_OUT: print("CameraGetImageBuffer failed({}): {}".format(e.error_code, e.message)) # 关闭相机 mvsdk.CameraUnInit(hCamera) # 释放帧缓存 mvsdk.CameraAlignFree(pFrameBuffer) if __name__ == "__main__": # import os # os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" # os.environ["CUDA_VISIBLE_DEVICES"] = "-1" dataset = generate.DataSet("/home/xinyang/Workspace/box_resize") train(dataset, show_bar=True) input("press enter to continue...")