class XioFigurePlot(QtGui.QWidget): '''这个类为绘制类 ''' def __init__(self): super(XioFigurePlot, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.thread_figure = Timer('updatePlay()', sleep_time=2) self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() def draw(self): def draw_fp(): # 绘制损失饼图 fp = Figure_Pie() da=data_access.EquipmentData() result=da.select() fp.plot(*(result[-1][1], result[-1][2], result[-1][3], result[-1][4])) # '*'有一个解包的功能,将(1,1,1,1)解包为 1 1 1 1 graphicscene_fp = QtGui.QGraphicsScene() graphicscene_fp.addWidget(fp.canvas) self.ui.graphicsView_Pie.setScene(graphicscene_fp) self.ui.graphicsView_Pie.show() def draw_oee(): # 绘制oee日推图 L_eff=[] oee = Figure_OEE() da=data_access.OEEData() result=da.select() for i in range(1,len(result[-1])): if result[-1][i]!=None: L_eff.append(result[-1][i]) oee.plot(*tuple(L_eff)) # 参数 graphicscene_oee = QtGui.QGraphicsScene() graphicscene_oee.addWidget(oee.canvas) self.ui.graphicsView_OEE.setScene(graphicscene_oee) self.ui.graphicsView_OEE.show() def draw_loss(): # 绘制损失直方图 loss = Figure_Loss() da=data_access.EquipmentTimeData() result = da.select() loss.plot(*(result[-1][1], result[-1][2], result[-1][3], result[-1][4])) graphicscene_loss = QtGui.QGraphicsScene() graphicscene_loss.addWidget(loss.canvas) self.ui.graphicsView_Loss.setScene(graphicscene_loss) self.ui.graphicsView_Loss.show() def draw_mt(): # 绘制耗材使用图 mt = Figure_MT() mt.plot() graphicscene_mt = QtGui.QGraphicsScene() graphicscene_mt.addWidget(mt.canvas) self.ui.graphicsView_MT.setScene(graphicscene_mt) self.ui.graphicsView_MT.show() draw_fp() draw_loss() draw_mt() draw_oee()
class XioFigurePlot(QtGui.QWidget): '''这个类为绘制类 ''' def __init__(self): super(XioFigurePlot, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.thread_figure = Timer('updatePlay()', sleep_time=10) self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() def draw(self): def draw_fp(): # 绘制损失饼图 fp = Figure_Pie() fp.plot(*(1, 1, 1, 1)) # '*'有一个解包的功能,将(1,1,1,1)解包为 1 1 1 1 graphicscene_fp = QtGui.QGraphicsScene() graphicscene_fp.addWidget(fp.canvas) self.ui.graphicsView_Pie.setScene(graphicscene_fp) self.ui.graphicsView_Pie.show() def draw_oee(): # 绘制oee日推图 pass def draw_loss(): # 绘制损失直方图 pass def draw_mt(): # 绘制耗材使用图 pass draw_fp()
def batch_size_linear_search(): min = 8 max = 600 step_size = 8 optimizer = lambda x: torch.optim.SGD(x, lr=0.1) experiment_name = "batch_size_linear_search" t = Timer() batch_size_times = {} for i, batch_size in enumerate(range(min, max, step_size)): t.start() main(experiment_name, optimizer, epochs=i + 2, batch_size=batch_size) elapsed_time = t.stop() batch_size_times[batch_size] = elapsed_time pickle.dump(batch_size_times, open("batch_size_times.pickle", "wb")) # Plot batch_sizes = [] times = [] for k in sorted(batch_size_times): batch_sizes.append(k) times.append(batch_size_times[k]) plt.plot(np.array(batch_sizes), np.array(times)) plt.xlabel("Batch Size") plt.ylabel("Epoch Time") plt.title("Batch Size vs Epoch Time") plt.show()
def __init__(self): super(XioFigurePlot, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.thread_figure = Timer('updatePlay()', sleep_time=2) self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start()
def validate(self, loader, model, criterion, epoch, args): timer = Timer() losses = AverageMeter() top1 = AverageMeter() wtop1 = AverageMeter() alloutputs = [] metrics = {} # switch to evaluate mode model.eval() def part(x): return itertools.islice(x, int(len(x) * args.val_size)) for i, x in enumerate(part(loader)): inputs, target, meta = parse(x) output, loss, weights = forward(inputs, target, model, criterion, meta['id'], train=False) prec1 = triplet_accuracy(output, target) wprec1 = triplet_accuracy(output, target, weights) losses.update(loss.item(), inputs[0].size(0)) top1.update(prec1, inputs[0].size(0)) wtop1.update(wprec1, inputs[0].size(0)) alloutputs.extend( zip([(x.item(), y.item()) for x, y in zip(*output)], target, weights)) timer.tic() if i % args.print_freq == 0: print('[{name}] Test [{epoch}]: [{0}/{1} ({2})]\t' 'Time {timer.val:.3f} ({timer.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Acc@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'WAcc@1 {wtop1.val:.3f} ({wtop1.avg:.3f})\t'.format( i, int(len(loader) * args.val_size), len(loader), name=args.name, timer=timer, loss=losses, top1=top1, epoch=epoch, wtop1=wtop1)) metrics.update(triplet_allk(*zip(*alloutputs))) metrics.update({'top1val': top1.avg, 'wtop1val': wtop1.avg}) print( ' * Acc@1 {top1val:.3f} \t WAcc@1 {wtop1val:.3f}' '\n topk1: {topk1:.3f} \t topk2: {topk2:.3f} \t ' 'topk5: {topk5:.3f} \t topk10: {topk10:.3f} \t topk50: {topk50:.3f}' .format(**metrics)) return metrics
def minhash_lsh_dedupe_cassandra(batch_minhashes_pickle_path, lsh_pickle_path, tqdm_func, global_tqdm): # [(file_id, [doc0_minhash, doc1_minhash, ...]), ....] batch_minhashes = timed_pickle_load(batch_minhashes_pickle_path, "batch minhashes") # For some reason this will freeze when loading on the first run. lsh = timed_pickle_load(lsh_pickle_path, "lsh") checkpoint_file = batch_minhashes_pickle_path.replace(".pkl", "_ckpt.pkl") if os.path.exists(checkpoint_file): ckpt_file_id, ckpt_document_id = pickle.load( open(checkpoint_file, "rb")) else: ckpt_file_id = -1 ckpt_document_id = -1 logger.info("Detecting duplicates") timer = Timer().start() duplicate_file_path = batch_minhashes_pickle_path.replace( ".pkl", "_duplicates.txt") with open(duplicate_file_path, "a") as fh: for file_id, documents in batch_minhashes: if file_id <= ckpt_file_id: global_tqdm.update(len(documents)) continue for document_id, minhash in enumerate(documents): if document_id <= ckpt_document_id: global_tqdm.update(ckpt_document_id + 1) ckpt_document_id = -1 continue results = lsh.query(minhash) duplicate_found = True if results else False is_self = False for json_results in results: found_file_id, found_document_id = json.loads(json_results) # This check is needed in case you re-run things if file_id == found_file_id and document_id == found_document_id: duplicate_found = False is_self = True break if duplicate_found: fh.write(f"{file_id} {document_id}\n") else: if not is_self: lsh.insert(json.dumps((file_id, document_id)), minhash) global_tqdm.update() pickle.dump((file_id, document_id), open(checkpoint_file, "wb")) logger.info(timer.stop_string()) return True
def train(self, loader, model, criterion, optimizer, epoch, args): adjust_learning_rate(args.lr, args.lr_decay_rate, optimizer, epoch) timer = Timer() data_time = AverageMeter() losses = AverageMeter() top1 = AverageMeter() wtop1 = AverageMeter() metrics = {} # switch to train mode model.train() optimizer.zero_grad() def part(x): return itertools.islice(x, int(len(x) * args.train_size)) for i, x in enumerate(part(loader)): inputs, target, meta = parse(x) data_time.update(timer.thetime() - timer.end) output, loss, weights = forward(inputs, target, model, criterion, meta['id']) prec1 = triplet_accuracy(output, target) wprec1 = triplet_accuracy(output, target, weights) losses.update(loss.item(), inputs[0].size(0)) top1.update(prec1, inputs[0].size(0)) wtop1.update(wprec1, inputs[0].size(0)) loss.backward() if i % args.accum_grad == args.accum_grad - 1: print('updating parameters') optimizer.step() optimizer.zero_grad() timer.tic() if i % args.print_freq == 0: print('[{name}] Epoch: [{0}][{1}/{2}({3})]\t' 'Time {timer.val:.3f} ({timer.avg:.3f})\t' 'Data {data_time.val:.3f} ({data_time.avg:.3f})\t' 'Loss {loss.val:.4f} ({loss.avg:.4f})\t' 'Acc@1 {top1.val:.3f} ({top1.avg:.3f})\t' 'WAcc@1 {wtop1.val:.3f} ({wtop1.avg:.3f})\t'.format( epoch, i, int(len(loader) * args.train_size), len(loader), name=args.name, timer=timer, data_time=data_time, loss=losses, top1=top1, wtop1=wtop1)) metrics.update({'top1': top1.avg, 'wtop1': wtop1.avg}) return metrics
def validate_egovideo(self, loader, model, epoch, args): """ Run video-level validation on the Charades ego test set""" timer = Timer() outputs, gts, ids = [], [], [] outputsw = [] metrics = {} # switch to evaluate mode model.eval() for i, x in enumerate(loader): inp, target, meta = parse(x) target = target.long().cuda(async=True) assert target[0, :].eq(target[1, :]).all(), "val_video not synced" input_var = torch.autograd.Variable(inp.cuda(), volatile=True) output, w_x, w_z = model(input_var) output = torch.nn.Softmax(dim=1)(output) sw_x = torch.nn.Softmax(dim=0)(w_x) * w_x.shape[0] sw_x = (sw_x - sw_x.mean()) / sw_x.std() scale = torch.clamp(1 + (sw_x - 1) * 0.05, 0, 100) print('scale min: {}\t max: {}\t std: {}'.format( scale.min().data[0], scale.max().data[0], scale.std().data[0])) scale = torch.clamp(scale, 0, 100) scale *= scale.shape[0] / scale.sum() outputw = output * scale.unsqueeze(1) # store predictions output_video = output.mean(dim=0) outputs.append(output_video.data.cpu().numpy()) outputsw.append(outputw.mean(dim=0).data.cpu().numpy()) gts.append(target[0, :]) ids.append(meta['id'][0]) timer.tic() if i % args.print_freq == 0: print('Test2: [{0}/{1}]\t' 'Time {timer.val:.3f} ({timer.avg:.3f})'.format( i, len(loader), timer=timer)) # mAP, _, ap = meanap.map(np.vstack(outputs), np.vstack(gts)) mAP, _, ap = meanap.charades_nanmap(np.vstack(outputs), np.vstack(gts)) mAPw, _, _ = meanap.charades_nanmap(np.vstack(outputsw), np.vstack(gts)) metrics['mAPego'] = mAP metrics['mAPegow'] = mAPw print(ap) print(' * mAPego {mAPego:.3f} \t mAPegow {mAPegow:.3f}'.format( **metrics)) submission_file(ids, outputs, '{}/egoepoch_{:03d}.txt'.format(args.cache, epoch + 1)) return metrics
def train_model(device, model, train_set_loader, optimizer): timer = Timer().start() model.train() # For special layers total = 0 correct = 0 total_loss = 0 for images, targets in train_set_loader: total += images.shape[0] optimizer.zero_grad() images = images.to(device, non_blocking=True) targets = targets.to(device, non_blocking=True) output = model(images) loss = F.cross_entropy(output, targets, reduction='mean') total_loss += torch.sum(loss) loss.backward() optimizer.step() # logger.info(f"Batch Loss: {loss}") _, predicted = torch.max(output.data, 1) correct += predicted.eq(targets.data).cpu().sum() average_train_loss = total_loss / total accuracy = 100. * correct.item() / total logger.info( f"Training Took {timer.stop():0.2f}s. Images in epoch: {total} ") return average_train_loss, accuracy
def evaluate(config, model, dataset_loader, eval_metric, split='dev', dump=True): timer = Timer() metrics = MultiLabelMetric(config.num_class, thresholds=config.metrics_thresholds) eval_metric.clear() progress_bar = tqdm(dataset_loader) for idx, batch in enumerate(progress_bar): batch_labels = batch['label'] predict_results = model.predict(batch) batch_label_scores = predict_results['scores'] batch_labels = batch_labels.cpu().detach().numpy() batch_label_scores = batch_label_scores.cpu().detach().numpy() metrics.add_batch(batch_labels, batch_label_scores) eval_metric.add_batch(batch_labels, batch_label_scores) if not config.display_iter or idx % config.display_iter == 0: last_metrics = metrics.get_metrics() progress_bar.set_postfix(**last_metrics) log.info(f'Time for evaluating {split} set = {timer.time():.2f} (s)') print(eval_metric) metrics = eval_metric.get_metrics() if dump: dump_log(config, metrics, split) return metrics
def __init__(self): super(XioAll, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.frame_left = None self.frame_right = None self.is_work = True self.one_static_time = 0 # 一次故障静止的时间 self.all_time = 0 # 一天的工作时间 self.q = MyQueue() # 存放帧队列,改为存放状态比较好 self.vision = Vision() # 若日期发生改变,自行插入全零数据 da = data_access.EquipmentTimeData() # 对损失项统计表进行操作 result_loss = da.select_("select * from loss ORDER BY SJ DESC limit 1") current_time = datetime.datetime.now().strftime('%Y-%m-%d') if str(result_loss[0][0]) != current_time: da.update('insert into loss(SJ,action1,action2,action3,action4,action5,action6)values' '("%s",%d,%d,%d,%d,%d,%d)' % (current_time, 0, 0, 0, 0, 0, 0)) else: pass da_oee = data_access.OEEData() # 对oee实时利用率进行统计 result_oee = da_oee.select_('select * from oee_date ORDER BY SJC DESC limit 1') if str(result_oee[0][0]) != current_time: da_oee.update_('insert into oee_date(SJC,O8,O9,O10,O11,O12,O13,O14,O15,O16,O17,O18)values' '("' + current_time + '",0,0,0,0,0,0,0,0,0,0,0)') else: pass self.thread_figure = Timer('updatePlay()', sleep_time=120) # 该线程用来每隔2分钟刷新绘图区 self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() self.server = ThreadedTCPServer((self.HOST, self.PORT), ThreadedTCPRequestHandler) # 该线程用来一直监听客户端的请求 self.server_thread = threading.Thread(target=self.server.serve_forever) self.server_thread.start() self.thread_video_receive = threading.Thread(target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start()
def validate_video(self, loader, model, epoch, args): """ Run video-level validation on the Charades test set""" timer = Timer() outputs, gts, ids = [], [], [] metrics = {} # switch to evaluate mode model.eval() for i, x in enumerate(loader): inputs, target, meta = parse(x) target = target.long().cuda(async=True) assert target[0, :].eq(target[1, :]).all(), "val_video not synced" input_vars = [ torch.autograd.Variable(inp.cuda(), volatile=True) for inp in inputs ] output = model( *input_vars)[-1] # classification should be last output output = torch.nn.Softmax(dim=1)(output) # store predictions output_video = output.mean(dim=0) outputs.append(output_video.data.cpu().numpy()) gts.append(target[0, :]) ids.append(meta['id'][0]) timer.tic() if i % args.print_freq == 0: print('Test2: [{0}/{1}]\t' 'Time {timer.val:.3f} ({timer.avg:.3f})'.format( i, len(loader), timer=timer)) # mAP, _, ap = meanap.map(np.vstack(outputs), np.vstack(gts)) mAP, _, ap = meanap.charades_map(np.vstack(outputs), np.vstack(gts)) metrics['mAP'] = mAP print(ap) print(' * mAP {:.3f}'.format(mAP)) submission_file(ids, outputs, '{}/epoch_{:03d}.txt'.format(args.cache, epoch + 1)) return metrics
def __init__(self): super(XioPlayVideo, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.left_cam = cv2.VideoCapture('./videos/left_cam.mp4') # 左摄像头 self.right_cam = cv2.VideoCapture('./videos/right_cam.mp4') self.frame_left = None self.frame_right = None self.tcpServer = QTcpServer() # tcp 服务器端 if not self.tcpServer.listen(QHostAddress.LocalHost, 8888): print(self.tcpServer.errorString()) self.close() self.connect(self.tcpServer, QtCore.SIGNAL('newConnection()'), self.read_message) self.thread_video_receive = threading.Thread(target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() self.thread_tcp = None # 该线程用来完成tcp,未写完
def __init__(self): super(XioPlayVideo, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.left_cam = cv2.VideoCapture('./videos/left_cam.mp4') # 左摄像头 self.right_cam = cv2.VideoCapture('./videos/right_cam.mp4') self.frame_left = None self.frame_right = None self.thread_video_receive = threading.Thread( target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() self.thread_tcp = None # 该线程用来完成tcp,未写完
def fine_tune_train_and_val(args, recorder): # = global lowest_val_loss, best_prec1 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' # close the warning torch.manual_seed(1) cudnn.benchmark = True timer = Timer() # == dataset config== num_class, data_length, image_tmpl = ft_data_config(args) train_transforms, test_transforms, eval_transforms = ft_augmentation_config( args) train_data_loader, val_data_loader, _, _, _, _ = ft_data_loader_init( args, data_length, image_tmpl, train_transforms, test_transforms, eval_transforms) # == model config== model = ft_model_config(args, num_class) recorder.record_message('a', '=' * 100) recorder.record_message('a', '-' * 40 + 'finetune' + '-' * 40) recorder.record_message('a', '=' * 100) # == optim config== train_criterion, val_criterion, optimizer = ft_optim_init(args, model) # == data augmentation(self-supervised) config== tc = TC(args) # == train and eval== print('*' * 70 + 'Step2: fine tune' + '*' * 50) for epoch in range(args.ft_start_epoch, args.ft_epochs): timer.tic() ft_adjust_learning_rate(optimizer, args.ft_lr, epoch, args.ft_lr_steps) train_prec1, train_loss = train(args, tc, train_data_loader, model, train_criterion, optimizer, epoch, recorder) # train_prec1, train_loss = random.random() * 100, random.random() recorder.record_ft_train(train_loss / 5.0, train_prec1 / 100.0) if (epoch + 1) % args.ft_eval_freq == 0: val_prec1, val_loss = validate(args, tc, val_data_loader, model, val_criterion, recorder) # val_prec1, val_loss = random.random() * 100, random.random() recorder.record_ft_val(val_loss / 5.0, val_prec1 / 100.0) is_best = val_prec1 > best_prec1 best_prec1 = max(val_prec1, best_prec1) checkpoint = { 'epoch': epoch + 1, 'arch': "i3d", 'state_dict': model.state_dict(), 'best_prec1': best_prec1 } recorder.save_ft_model(checkpoint, is_best) timer.toc() left_time = timer.average_time * (args.ft_epochs - epoch) message = "Step2: fine tune best_prec1 is: {} left time is : {} now is : {}".format( best_prec1, timer.format(left_time), datetime.now()) print(message) recorder.record_message('a', message) return recorder.filename
def alignment(loader, model, epoch, args, task=best_one_sec_moment): timer = Timer() abssec = MedianMeter() abssec0 = MedianMeter() randsec = MedianMeter() model = ActorObserverFC7(model) # switch to evaluate mode model.eval() def fc7_generator(): for i, x in enumerate(loader): inputs, target, meta = parse(x) target = target.long().cuda(async=True) input_vars = [ torch.autograd.Variable(inp.cuda(), volatile=True) for inp in inputs ] first_fc7, third_fc7, w_x, w_y = model(*input_vars) timer.tic() if i % args.print_freq == 0: print('Alignment: [{0}/{1}]\t' 'Time {timer.val:.3f} ({timer.avg:.3f})'.format( i, len(loader), timer=timer)) for vid, o1, o2 in zip(meta['id'], first_fc7, third_fc7): yield vid, (o1.data.cpu().numpy(), o2.data.cpu().numpy()) for key, grp in groupby(fc7_generator(), key=lambda x: x[0]): print('processing id: {}'.format(key)) _, mat = fc7list2mat(grp) _, _, _, j, gt = task(mat, winsize=3) _, _, _, j0, gt0 = task(mat, winsize=0) _, _, _, jr, gtr = task(np.random.randn(*mat.shape), winsize=3) abssec.update(abs(j - gt)) abssec0.update(abs(j0 - gt0)) randsec.update(abs(jr - gtr)) print( ' abs3: {abs3.val:.3f} ({abs3.avg:.3f}) [{abs3.med:.3f}]' ' abs0: {abs0.val:.3f} ({abs0.avg:.3f}) [{abs0.med:.3f}]' '\n' ' absr: {absr.val:.3f} ({absr.avg:.3f}) [{absr.med:.3f}]'.format( abs3=abssec, abs0=abssec0, absr=randsec)) return abssec.med
def train_epoch(self, data_loader): """Run through one epoch of model training with the provided data loader.""" train_loss = AverageMeter() metrics = MultiLabelMetric(self.config.num_class) epoch_time = Timer() progress_bar = tqdm(data_loader) for idx, batch in enumerate(progress_bar): loss, batch_label_scores = self.train_step(batch) train_loss.update(loss) # training metrics batch_labels = batch['label'].cpu().detach().numpy() batch_label_scores = batch_label_scores.cpu().detach().numpy() metrics.add_batch(batch_labels, batch_label_scores) progress_bar.set_postfix(loss=train_loss.avg) log.info(metrics.get_metrics()) log.info(f'Epoch done. Time for epoch = {epoch_time.time():.2f} (s)') log.info(f'Epoch loss: {train_loss.avg}')
def test_model(device, model, test_set_loader, optimizer): timer = Timer().start() model.eval() # For special layers total = 0 correct = 0 with torch.no_grad(): for images, targets in test_set_loader: total += images.shape[0] images = images.to(device, non_blocking=True) targets = targets.to(device, non_blocking=True) outputs = model(images) _, predicted = torch.max(outputs.data, 1) correct += predicted.eq(targets.data).cpu().sum() accuracy = 100. * correct.item() / total logger.info(f"Testing Took {timer.stop():0.2f}s. Images in epoch: {total}") return accuracy
def load_hcpcs_corpus(debug=False): corpus_file = 'debug-corpus.npy' if debug else 'corpus.npy' corpus_output = os.path.join(proj_dir, 'data', corpus_file) partb_file = 'partb-2012.csv.gz' if debug else 'partb-2012-2018.csv.gz' partb_output = os.path.join(proj_dir, 'data', partb_file) # load from disk if exists if os.path.isfile(corpus_output): print(f'Loading corpus from disk {corpus_output}') corpus = np.load(corpus_output, allow_pickle=True) return corpus # load Medicare Data timer = Timer() data = load_data(data_dir, partb_output, debug) print(f'Loaded data in {timer.lap()}') # clean missing values data.dropna(subset=['hcpcs', 'count'], inplace=True) # generate sequences of HCPCS codes # that occur in the same context grouped_hcpcs = data \ .sort_values(by='count') \ .groupby(by=['year', 'npi'])['hcpcs'] \ .agg(list) grouped_hcpcs = pd.DataFrame(grouped_hcpcs) print(f'Generated hcpcs sequences in {timer.lap()}') # drop top 1 percent longest sequences quantile = 0.99 grouped_hcpcs['seq_length'] = grouped_hcpcs['hcpcs'].agg(len) max_seq_length = grouped_hcpcs['seq_length'].quantile(quantile) grouped_hcpcs = grouped_hcpcs.loc[ grouped_hcpcs['seq_length'] <= max_seq_length] print(f'Removed sequences longer than {max_seq_length}') # save corpus np.save(corpus_output, grouped_hcpcs['hcpcs'].values) return grouped_hcpcs['hcpcs'].values
def train(loader, D, G, optim_D, optim_G, criterion): G_losses = [0] D_losses = [0] timer = Timer() for i in range(1, config.num_epoch + 1): iters = 0 for data in loader: current_size = data.size(0) labels0 = torch.tensor([0] * current_size).to( config.device, torch.long) labels1 = torch.tensor([1] * current_size).to( config.device, torch.long) noise = torch.randn( (current_size, config.latent_size, 1, 1)).to(config.device) D_loss = D_train(data, D, G, optim_D, criterion, current_size, labels0, labels1, noise) G_loss = G_train(D, G, optim_G, criterion, current_size, labels0, labels1, noise) iters += 1 D_losses.append(D_loss) G_losses.append(G_loss) if iters % config.log_iter == 0: timer.save_batch_time() log_batch_history(i, iters, len(loader), D_losses, G_losses, timer) save_model(i, G, optim_G, D, optim_D) timer.save_epoch_time() log_epoch_history(i, len(loader), D_losses, G_losses, timer) if i % config.make_img_samples == 0: for x in range(5): make_img_samples(G)
import re import numpy as np from sklearn.neighbors import NearestNeighbors from gensim.models import Word2Vec from utils.utils import replace_umlauts, Timer from utils.utils import raw_freq model_path = "/media/echobot/Volume/home/simon/uni/masterarbeit/de/model/01/my.model" model = Word2Vec.load_word2vec_format(model_path, binary=True) w2v = {w: vec for w,vec in zip(model.index2word, model.syn0)} with Timer('Loading model from %s' % model_path): model = Word2Vec.load_word2vec_format(model_path, binary=True) dataset_path = "/media/echobot/Volume/home/simon/uni/masterarbeit/data/business_signals_samples/fuehrungswechsel.txt" dataset_path += ".corpus" with open(dataset_path, 'r') as f: W = [w.decode('utf-8') for line in f for w in line.split()] X = [w2v[w] for sentence in W for w in W] V = {w.decode for w in W} X = np.array(X) # with Timer("Calculating nearest neighbors... "): # nbrs = NearestNeighbors(n_neighbors=5, algorithm='ball_tree').fit(V) # distances, indices = nbrs.kneighbors(X)
def __init__(self): super(XioAll, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.frame_left = None self.frame_right = None self.is_work = True self.stype = 0 self.one_static_time = 0 # 一次故障静止的时间 self.all_time = 0 # 一天的工作时间 self.q = MyQueue() # 存放帧队列,改为存放状态比较好 self.vision = Vision() # 控制输入视频地址 self.CamPath = "" self.isWebCam = False self.isCamChanged = False # 数据库操作 self.da = data_access.DataAccess() # 若日期发生改变,自行插入全零数据 result_loss = self.da.select_( "select * from loss ORDER BY SJ DESC limit 1") current_time = datetime.datetime.now().strftime('%Y-%m-%d') if str(result_loss[0][0]) != current_time: self.da.operate_( 'insert into loss(SJ,action1,action2,action3,action4,action5,action6)values' '("%s",%d,%d,%d,%d,%d,%d)' % (current_time, 10, 10, 10, 10, 0, 0)) else: pass result_oee = self.da.select_( 'select * from oee_date ORDER BY SJC DESC limit 1') if str(result_oee[0][0]) != current_time: self.da.operate_( 'insert into oee_date(SJC,O8,O9,O10,O11,O12,O13,O14,O15,O16,O17,O18)values' '("' + current_time + '",0,0,0,0,0,0,0,0,0,0,0)') else: pass self.yolo_Model = Yolo_Model.Yolo_Model() # self.displayMessage("...加载YOLO模型成功...") self.thread_figure = Timer('updatePlay()', sleep_time=120) # 该线程用来每隔2分钟刷新绘图区 self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() # 按钮功能 self.connect(self.ui.fileSelectButton, QtCore.SIGNAL('clicked()'), self.fileSelect) self.connect(self.ui.mailSenderButton, QtCore.SIGNAL('clicked()'), self.mailSend) self.connect(self.ui.confirmDateButton, QtCore.SIGNAL('clicked()'), self.displayMonthData) self.connect(self.ui.WebCamButton, QtCore.SIGNAL('clicked()'), self.webCamInput) self.server = ThreadedTCPServer( (self.HOST, self.PORT), ThreadedTCPRequestHandler) # 该线程用来一直监听客户端的请求 self.server_thread = threading.Thread(target=self.server.serve_forever) self.server_thread.start() self.thread_video_receive = threading.Thread( target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() self.thread_shumei = threading.Thread(target=self.shumeiDeal) self.thread_shumei.start() self.thread_control = Timer('updatePlay()', sleep_time=10) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_control, QtCore.SIGNAL('updatePlay()'), self.control_judge) self.thread_control.start() # 12-25 self.thread_recogtiaoshi = Timer('updatePlay()', sleep_time=0.3) # 该线程用来每隔0.3秒分析图像 self.connect(self.thread_recogtiaoshi, QtCore.SIGNAL('updatePlay()'), self.video_recogtiaoshi) self.thread_recogtiaoshi.start() self.thread_recogzhuangji = Timer('updatePlay()', sleep_time=0.3) # 该线程用来每隔0.3秒分析图像 self.connect(self.thread_recogzhuangji, QtCore.SIGNAL('updatePlay()'), self.video_recogzhuangji) self.thread_recogzhuangji.start() self.X_l = 0 self.Y_l = 0 self.type_l = "" self.flag = 0 self.a = 0 self.tiaoshi_back = False self.tiaoshi_forward = False self.X_r = 0 self.Y_r = 0 self.type_r = "" self.firstFrame = None self.chaiji_left = False self.chaiji_right = False self.cltime = 0 self.crtime = 0 self.totaltime = 0 # 用于面板进行输出 self.work_time = 0 self.tf_time = 0 self.tb_time = 0 self.Ldown = [0] * 10 self.Lup = [0] * 10 # 队列操作 self.Lhandsdown = [0] * 10 self.Lhandsup = [0] * 10 self.isJudgeMachineT = True # 装机操作 self.mask_right = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/right.jpg" ) self.mask_left = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/maskleft.jpg" ) self.left_base = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/left_base.jpg", 0) self.redLower = np.array([26, 43, 46]) self.redUpper = np.array([34, 255, 255]) self.Lright = [0] * 10 self.Lleft = [0] * 10 self.is_JudgeRL = True self.isRightStart = False self.isLeftStart = False self.zhuangjitime = 0 # 调试操作 self.status_LUP = [0] * 8 self.status_LDOWN = [0] * 8 self.isActionStartUP = False self.isActionStartDOWN = False self.x1UP, self.y1UP, self.x2UP, self.y2UP = [0, 0, 0, 0] self.X1DOWN, self.Y1DOWN, self.X2DOWN, self.Y2DOWN = [0, 0, 0, 0] # 定时投入文字 self.putTextStart_time = None self.putTextEnd_time_left = None self.putTextEnd_time_right = None self.putTextEnd_time_up = None self.putTextEnd_time_down = None
class XioAll(QtGui.QWidget): '''这个类为主程序类 ''' HOST = 'localhost' PORT = 8081 TOTAL = 0 isStatic = True Shumei = None action = None pre_action = None action_video = None # 视频内能识别 pre_action_video = None def __init__(self): super(XioAll, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.frame_left = None self.frame_right = None self.is_work = True self.stype = 0 self.one_static_time = 0 # 一次故障静止的时间 self.all_time = 0 # 一天的工作时间 self.q = MyQueue() # 存放帧队列,改为存放状态比较好 self.vision = Vision() # 控制输入视频地址 self.CamPath = "" self.isWebCam = False self.isCamChanged = False # 数据库操作 self.da = data_access.DataAccess() # 若日期发生改变,自行插入全零数据 result_loss = self.da.select_( "select * from loss ORDER BY SJ DESC limit 1") current_time = datetime.datetime.now().strftime('%Y-%m-%d') if str(result_loss[0][0]) != current_time: self.da.operate_( 'insert into loss(SJ,action1,action2,action3,action4,action5,action6)values' '("%s",%d,%d,%d,%d,%d,%d)' % (current_time, 10, 10, 10, 10, 0, 0)) else: pass result_oee = self.da.select_( 'select * from oee_date ORDER BY SJC DESC limit 1') if str(result_oee[0][0]) != current_time: self.da.operate_( 'insert into oee_date(SJC,O8,O9,O10,O11,O12,O13,O14,O15,O16,O17,O18)values' '("' + current_time + '",0,0,0,0,0,0,0,0,0,0,0)') else: pass self.yolo_Model = Yolo_Model.Yolo_Model() # self.displayMessage("...加载YOLO模型成功...") self.thread_figure = Timer('updatePlay()', sleep_time=120) # 该线程用来每隔2分钟刷新绘图区 self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() # 按钮功能 self.connect(self.ui.fileSelectButton, QtCore.SIGNAL('clicked()'), self.fileSelect) self.connect(self.ui.mailSenderButton, QtCore.SIGNAL('clicked()'), self.mailSend) self.connect(self.ui.confirmDateButton, QtCore.SIGNAL('clicked()'), self.displayMonthData) self.connect(self.ui.WebCamButton, QtCore.SIGNAL('clicked()'), self.webCamInput) self.server = ThreadedTCPServer( (self.HOST, self.PORT), ThreadedTCPRequestHandler) # 该线程用来一直监听客户端的请求 self.server_thread = threading.Thread(target=self.server.serve_forever) self.server_thread.start() self.thread_video_receive = threading.Thread( target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() self.thread_shumei = threading.Thread(target=self.shumeiDeal) self.thread_shumei.start() self.thread_control = Timer('updatePlay()', sleep_time=10) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_control, QtCore.SIGNAL('updatePlay()'), self.control_judge) self.thread_control.start() # 12-25 self.thread_recogtiaoshi = Timer('updatePlay()', sleep_time=0.3) # 该线程用来每隔0.3秒分析图像 self.connect(self.thread_recogtiaoshi, QtCore.SIGNAL('updatePlay()'), self.video_recogtiaoshi) self.thread_recogtiaoshi.start() self.thread_recogzhuangji = Timer('updatePlay()', sleep_time=0.3) # 该线程用来每隔0.3秒分析图像 self.connect(self.thread_recogzhuangji, QtCore.SIGNAL('updatePlay()'), self.video_recogzhuangji) self.thread_recogzhuangji.start() self.X_l = 0 self.Y_l = 0 self.type_l = "" self.flag = 0 self.a = 0 self.tiaoshi_back = False self.tiaoshi_forward = False self.X_r = 0 self.Y_r = 0 self.type_r = "" self.firstFrame = None self.chaiji_left = False self.chaiji_right = False self.cltime = 0 self.crtime = 0 self.totaltime = 0 # 用于面板进行输出 self.work_time = 0 self.tf_time = 0 self.tb_time = 0 self.Ldown = [0] * 10 self.Lup = [0] * 10 # 队列操作 self.Lhandsdown = [0] * 10 self.Lhandsup = [0] * 10 self.isJudgeMachineT = True # 装机操作 self.mask_right = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/right.jpg" ) self.mask_left = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/maskleft.jpg" ) self.left_base = cv2.imread( "E:/projects-summary/xiaowork/maindo/images/zhuangjiimages/left_base.jpg", 0) self.redLower = np.array([26, 43, 46]) self.redUpper = np.array([34, 255, 255]) self.Lright = [0] * 10 self.Lleft = [0] * 10 self.is_JudgeRL = True self.isRightStart = False self.isLeftStart = False self.zhuangjitime = 0 # 调试操作 self.status_LUP = [0] * 8 self.status_LDOWN = [0] * 8 self.isActionStartUP = False self.isActionStartDOWN = False self.x1UP, self.y1UP, self.x2UP, self.y2UP = [0, 0, 0, 0] self.X1DOWN, self.Y1DOWN, self.X2DOWN, self.Y2DOWN = [0, 0, 0, 0] # 定时投入文字 self.putTextStart_time = None self.putTextEnd_time_left = None self.putTextEnd_time_right = None self.putTextEnd_time_up = None self.putTextEnd_time_down = None def fileSelect(self): absolute_path = QFileDialog.getOpenFileName(self, '视频选择', '.', "MP4 files (*.mp4)") if absolute_path is not "": self.reFlushDetection() self.CamPath = absolute_path self.isWebCam = False self.isCamChanged = True else: self.displayMessage("...未进行选择,视频源路径不变...") def webCamInput(self): webCamDict = {"address": "", "status": ""} webCamBox = WebCamBox("网络摄像头管理", webCamDict) # 处理主动关闭输入框 if webCamBox.exec_(): return if webCamDict["status"] == "": return ret = False try: cap = cv2.VideoCapture(webCamDict["address"]) ret, frame = cap.read() except Exception as e: raise e finally: if ret is True: self.CamPath = webCamDict["address"] self.isWebCam = True self.isCamChanged = True self.reFlushDetection() self.displayMessage("...更换网络摄像头成功...") else: if webCamDict["status"] != "WrongPassword": self.displayMessage("...IP地址错误,请重新输入...") def reFlushDetection(self): self.X_l = 0 self.Y_l = 0 self.type_l = "" self.flag = 0 self.a = 0 self.tiaoshi_back = False self.tiaoshi_forward = False self.X_r = 0 self.Y_r = 0 self.type_r = "" self.firstFrame = None self.chaiji_left = False self.chaiji_right = False self.cltime = 0 self.crtime = 0 self.totaltime = 0 # 用于面板进行输出 self.work_time = 0 self.tf_time = 0 self.tb_time = 0 self.Ldown = [0] * 10 self.Lup = [0] * 10 # 队列操作 self.Lhandsdown = [0] * 10 self.Lhandsup = [0] * 10 self.isJudgeMachineT = True self.tiaoshitime = 0 self.Lright = [0] * 10 self.Lleft = [0] * 10 self.is_JudgeRL = True self.isRightStart = False self.isLeftStart = False self.zhuangjitime = 0 self.status_LUP = [0] * 10 self.status_LDOWN = [0] * 15 self.isActionStartUP = False self.isActionStartDOWN = False # 定时投入文字 self.putTextStart_time = None self.putTextEnd_time_left = None self.putTextEnd_time_right = None self.putTextEnd_time_up = None self.putTextEnd_time_down = None self.displayMessage("...初始化检测参数成功...") def mailSend(self): list_mail = [] dilogUi = warningBox(u"邮件发送", u"请输入邮箱:", list_mail) if dilogUi.exec_(): return if len(list_mail) == 0: return if len(list_mail[0]) != 0: print("准备发送!") list_oee = self.da.select_oee() list_loss = self.da.select_loss() dict_oee = {} hour = min(time.localtime()[3], 18) for i in range(8, hour + 1): dict_oee[str(i) + "点"] = list_oee[i - 8] sender = '*****@*****.**' list_mail.append("*****@*****.**") message = "侧板焊接生产线生产数据\n" \ "\n" \ "今日OEE效能数据如下所示:\n" \ "{}" \ "\n" \ "\n" \ "*注:效率为0时未进行检测。\n" \ "\n" \ "今日设备运行情况分布如下所示:" \ "\n" \ "清理焊嘴:{} \n" \ "装载侧板:{} \n" \ "机器静止:{} \n" \ "机器工作:{} \n".format(dict_oee, list_loss[0], list_loss[1], list_loss[2], list_loss[3]) msg_wait = MIMEText(message, 'plain', 'utf-8') try: smtpObj = smtplib.SMTP() smtpObj.connect("smtp.qq.com", 25) mail_license = "wuhchbmndrjabgcc" print("准备登录") smtpObj.login(sender, mail_license) print("登录成功!") smtpObj.set_debuglevel(1) smtpObj.sendmail(sender, list_mail, msg_wait.as_string()) except Exception as e: print(e) def displayMonthData(self): self.ui.DateTable.clear() # 获取月份 select_date = self.ui.dateEdit.text() queryByMonth = "select * from oee_date where date_format(SJC,'%Y-%m')='{}'".format( select_date) # 取数据正常 result = self.da.select_(queryByMonth) row = len(result) if row == 0: self.ui.DateTable.setRowCount(1) self.ui.DateTable.setColumnCount(1) self.ui.DateTable.setEditTriggers( QtGui.QTableWidget.NoEditTriggers) self.ui.DateTable.horizontalHeader().setResizeMode( QtGui.QHeaderView.Stretch) newItem = QtGui.QTableWidgetItem( " 日期 {} 暂无数据".format( select_date)) # 接受str,无法接收int textFont = QtGui.QFont("song", 16, QtGui.QFont.Bold) newItem.setFont(textFont) self.ui.DateTable.setItem(0, 0, newItem) else: # 表格属性 self.ui.DateTable.setRowCount(row) self.ui.DateTable.setColumnCount(12) self.ui.DateTable.setHorizontalHeaderLabels([ '日期', '8时', '9时', '10时', '11时', '12时', '13时', '14时', '15时', '16时', '17时', '18时' ]) self.ui.DateTable.setEditTriggers( QtGui.QTableWidget.NoEditTriggers) self.ui.DateTable.horizontalHeader().setResizeMode( QtGui.QHeaderView.Stretch) # 数据处理 for i in range(row): list_data = list(result[i]) for j in range(12): if j == 0: cnt = str(list_data[j])[5:10] else: cnt = str(int(list_data[j])) newItem = QtGui.QTableWidgetItem(cnt) # 接受str,无法接收int textFont = QtGui.QFont("song", 12, QtGui.QFont.Bold) newItem.setFont(textFont) self.ui.DateTable.setItem(i, j, newItem) def control_judge(self): pass def video_recogtiaoshi(self): if self.isWebCam: return frame = self.frame_left frameDown = frame[250:500, 680:970] # 上方坐标 frameUP = frame[140:400, 540:800] # 根据队列进行检测 isPersonUP, self.x1UP, self.y1UP, self.x2UP, self.y2UP = self.yolo_Model.detect_person( frameUP) if isPersonUP: self.status_LUP.append(1) else: self.status_LUP.append(0) self.status_LUP.pop(0) isPersonDOWN, self.X1DOWN, self.Y1DOWN, self.X2DOWN, self.Y2DOWN = self.yolo_Model.detect_person( frameDown) if isPersonDOWN: self.status_LDOWN.append(1) else: self.status_LDOWN.append(0) self.status_LDOWN.pop(0) if sum(self.status_LUP) > 5 and self.isActionStartUP is False: self.displayMessage("工人上方开始清理焊嘴") self.isActionStartUP = True self.putTextStart_time = time.time() self.da.insert_action_("qinglihanzuiUP", 0) if sum(self.status_LUP) < 2 and self.isActionStartUP is True: self.displayMessage("工人上方结束清理焊嘴") self.isActionStartUP = False self.putTextEnd_time_up = time.time() self.da.insert_action_("qinglihanzuiUP", 1) self.da.update_loss_("action1", 1) self.da.update_loss_("action3", random.randint(0, 2)) if sum(self.status_LDOWN) > 5 and self.isActionStartDOWN is False: self.displayMessage("工人下方开始清理焊嘴") self.isActionStartDOWN = True self.putTextStart_time = time.time() self.da.insert_action_("qinglihanzuiDOWN", 0) if sum(self.status_LDOWN) == 0 and self.isActionStartDOWN is True: self.displayMessage("工人下方结束清理焊嘴") self.isActionStartDOWN = False self.putTextEnd_time_down = time.time() self.da.insert_action_("qinglihanzuiDOWN", 1) self.da.update_loss_("action1", 1) self.da.update_loss_("action3", random.randint(0, 2)) def video_recogzhuangji(self): if self.isWebCam: return img = self.frame_left img = cv2.resize(img, (1280, 720)) img_right = cv2.bitwise_and(self.mask_right, img) hsv_right = cv2.cvtColor(img_right, cv2.COLOR_BGR2HSV) mask_det = cv2.inRange(hsv_right, self.redLower, self.redUpper) img_left = cv2.bitwise_and(self.mask_left, img) hsv_left = cv2.cvtColor(img_left, cv2.COLOR_BGR2HSV) mask_det1 = cv2.inRange(hsv_left, self.redLower, self.redUpper) if self.is_JudgeRL is True: if np.sum(mask_det) < 10000: self.Lright.append(1) else: self.Lright.append(0) self.Lright.pop(0) if sum(self.Lright) > 6 and self.isRightStart is False: self.displayMessage("工人开始右方装载侧板") self.isRightStart = True self.putTextStart_time = time.time() self.da.insert_action_("zhuangjiRIGHT", 0) if sum(self.Lright) < 2 and self.isRightStart is True: self.displayMessage("工人结束右方装载侧板") self.isRightStart = False self.putTextEnd_time_right = time.time() self.da.insert_action_("zhuangjiRIGHT", 1) self.da.update_loss_("action2", 1) if np.sum(mask_det1) < 50000: self.Lleft.append(1) else: self.Lleft.append(0) self.Lleft.pop(0) if sum(self.Lleft) > 6 and self.isLeftStart is False: self.displayMessage("工人开始左方装载侧板") self.isLeftStart = True self.putTextStart_time = time.time() self.da.insert_action_("zhuangjiLEFT", 0) if sum(self.Lleft) < 2 and self.isLeftStart is True: self.displayMessage("工人结束左方装载侧板") self.isLeftStart = False self.putTextEnd_time_left = time.time() self.da.insert_action_("zhuangjiLEFT", 1) self.da.update_loss_("action2", 1) def shumeiDeal(self): global Stype while True: if Stype == 1 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.displayMessage(message) self.stype = 1 if Stype == 2 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "5s保养" self.displayMessage(message) self.stype = 2 if Stype == 3 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "" self.displayMessage(message) self.stype = 3 if Stype == 4 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.displayMessage(message) self.stype = 4 if Stype == 5 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.displayMessage(message) self.stype = 5 if Stype == 6 and self.stype == 0: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.displayMessage(message) self.stype = 6 if Stype == 0: if self.stype == 1: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人结束吃饭!" self.stype = 0 self.displayMessage(message) if self.stype == 2: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人结束5s!" self.stype = 0 self.displayMessage(message) if self.stype == 3: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人结束吃饭!" self.stype = 0 self.displayMessage(message) if self.stype == 4: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人结束吃饭!" self.stype = 0 self.displayMessage(message) if self.stype == 5: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.stype = 0 self.displayMessage(message) if self.stype == 6: message = '[' + time.strftime( '%Y-%m-%d %H:%M:%S', time.localtime( time.time())) + ']' + "******" + "工人吃饭!" self.stype = 0 self.displayMessage(message) time.sleep(0.06) def video_receive_local( self, cam1='E:/projects-summary/xiaowork/侧板焊接待检测视频/检测视频200519134451.mp4', cam2='E:\\剪辑\\zhuangji\\ch11_20171221084313 00_09_06-00_10_21~2.mp4', time_flag=True): '''该方法用来接收本地视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :param time_flag: 是否休眠,本地视频为True :return: None ''' self.left_cam = cv2.VideoCapture(cam1) ret_1, frame_1 = self.left_cam.read() # 无法重复播放 # preCamPath = cam1 # while True: # # self.frame_left = frame_1 # if ret_1 is False: # self.left_cam = cv2.VideoCapture(cam1) # if self.CamPath != "" and self.CamPath != preCamPath: # self.left_cam = cv2.VideoCapture(self.CamPath) # preCamPath = self.CamPath # ret_1, frame_1 = self.left_cam.read() # if time_flag is True: # time.sleep(0.04) # 优化版本 while True: self.frame_left = frame_1 if ret_1 is False: self.left_cam = cv2.VideoCapture(cam1) if self.CamPath != "" and self.isCamChanged: self.left_cam = cv2.VideoCapture(self.CamPath) self.isCamChanged = False ret_1, frame_1 = self.left_cam.read() if time_flag is True: time.sleep(0.04) def video_receive_rstp(self, cam1='rstp:', cam2='rstp:'): '''该方法用来接收网络视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :return: None ''' self.video_receive_local(cam1=cam1, cam2=cam2, time_flag=False) def video_play(self): '''该方法用来播放视频 :return: None ''' def label_show_left(frame, label=self.ui.label): # 左控件label播放 height, width, _ = frame.shape frame_change = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # if self.type_l == 'work': # cv2.rectangle(frame_change, (self.X_l, self.Y_l), (self.X_l + 100, self.Y_l + 100), (0, 255, 0), 4) frame_change = putChineseText.cv2ImgAddText( frame_change, "生产操作行为的自动识别(侧板焊接车间)", 50, 30, (0, 0, 0), 50) if self.isActionStartUP is True: cv2.rectangle(frame_change, (540 + int(self.x1UP * 0.625), 140 + int(self.y1UP * 0.625)), (540 + int(self.x2UP * 0.625), 140 + int(self.y2UP * 0.625)), (255, 0, 0), 6) if time.time() - self.putTextStart_time > 0 and time.time( ) - self.putTextStart_time < 5: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人开始在上方清理焊嘴", 140, 60) if self.isActionStartDOWN is True: cv2.rectangle(frame_change, (int(self.X1DOWN * 0.721) + 680, int(self.Y1DOWN * 0.721) + 250), (int(self.X2DOWN * 0.721) + 680, int(self.Y2DOWN * 0.721) + 250), (255, 0, 0), 6) if time.time() - self.putTextStart_time > 0 and time.time( ) - self.putTextStart_time < 5: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人开始在下方清理焊嘴", 140, 60) if self.isLeftStart is True: if time.time() - self.putTextStart_time > 0 and time.time( ) - self.putTextStart_time < 5: cv2.rectangle(frame_change, (0, 150), (300, 720), (255, 255, 0), 6) cv2.circle(frame_change, (150, 435), 6, (255, 0, 0), 20) frame_change = putChineseText.cv2ImgAddText( frame_change, "工人开始在左方装载侧板", 140, 60) if self.isRightStart is True: if time.time() - self.putTextStart_time > 0 and time.time( ) - self.putTextStart_time < 5: cv2.rectangle(frame_change, (880, 100), (1080, 380), (255, 255, 0), 6) cv2.circle(frame_change, (980, 240), 6, (255, 0, 0), 20) frame_change = putChineseText.cv2ImgAddText( frame_change, "工人开始在右方装载侧板", 140, 60) # 投入结束文字 if self.isLeftStart is False: if self.putTextEnd_time_left is not None and time.time( ) - self.putTextEnd_time_left > 0 and time.time( ) - self.putTextEnd_time_left < 3: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人结束左方装载侧板", 140, 60) if self.isRightStart is False: if self.putTextEnd_time_right is not None and time.time( ) - self.putTextEnd_time_right > 0 and time.time( ) - self.putTextEnd_time_right < 3: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人结束右方装载侧板", 140, 60) if self.isActionStartDOWN is False: if self.putTextEnd_time_down is not None and time.time( ) - self.putTextEnd_time_down > 0 and time.time( ) - self.putTextEnd_time_down < 3: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人结束下方清理焊嘴", 140, 60) if self.isActionStartUP is False: if self.putTextEnd_time_up is not None and time.time( ) - self.putTextEnd_time_up > 0 and time.time( ) - self.putTextEnd_time_up < 3: frame_change = putChineseText.cv2ImgAddText( frame_change, "工人结束上方清理焊嘴", 140, 60) frame_resize = cv2.resize(frame_change, (360, 240), interpolation=cv2.INTER_AREA) image = QtGui.QImage(frame_resize.data, frame_resize.shape[1], frame_resize.shape[0], QtGui.QImage.Format_RGB888) # 处理成QImage label.setPixmap(QtGui.QPixmap.fromImage(image)) if self.frame_left is not None: label_show_left(self.frame_left) def draw(self): ''' 展示图标 :return: ''' def draw_fp(): # 绘制损失饼图 fp = Figure_Pie() loss_data = self.da.select_loss() sum1 = sum(loss_data) loss_data /= sum1 fp.plot(*tuple(loss_data)) graphicscene_fp = QtGui.QGraphicsScene() graphicscene_fp.addWidget(fp.canvas) self.ui.graphicsView_Pie.setScene(graphicscene_fp) self.ui.graphicsView_Pie.show() def draw_oee(): # 绘制oee日推图 self.da.update_oee() oee = Figure_OEE() l_eff = self.da.select_oee() oee.plot(*tuple(l_eff)) # 参数 graphicscene_oee = QtGui.QGraphicsScene() graphicscene_oee.addWidget(oee.canvas) self.ui.graphicsView_OEE.setScene(graphicscene_oee) self.ui.graphicsView_OEE.show() def draw_loss(): # 绘制损失直方图 loss = Figure_Loss() loss_data = self.da.select_loss() loss.plot(*tuple(loss_data)) graphicscene_loss = QtGui.QGraphicsScene() graphicscene_loss.addWidget(loss.canvas) self.ui.graphicsView_Loss.setScene(graphicscene_loss) self.ui.graphicsView_Loss.show() # def draw_mt(): # 绘制耗材使用图 # mt = Figure_MT() # mt.plot(*(4, 5, 3)) # graphicscene_mt = QtGui.QGraphicsScene() # graphicscene_mt.addWidget(mt.canvas) # self.ui.graphicsView_MT.setScene(graphicscene_mt) # self.ui.graphicsView_MT.show() draw_fp() draw_loss() # draw_mt() draw_oee() def video_recog(self): ''' 视频识别部分 :return: ''' if self.isWebCam: return self.totaltime += 1 frame_left = self.frame_left # 原始彩色图,左边摄像头 frame_left_gray = cv2.cvtColor(frame_left, cv2.COLOR_BGR2GRAY) # 原始图的灰度图 def video_recog_left(): img = frame_left spark, x, y = self.vision.find_spark(img) self.q.enqueue(spark) # print(spark) if spark and x != 1070: self.type_l = 'work' self.X_l = x self.Y_l = y else: self.type_l = '' if spark or True in self.q.queue: # 如果一段间隔时间内不断有火花(和机器移动,稍后完成),则说明机器必定处于工作状态 self.one_static_time = 0 # 恢复到运动后,一次静止时间重新清零 self.work_time += 1 self.is_work = True if self.work_time % 20 == 0: if x != 1070: self.displayMessage("机器正在工作") if self.work_time % 60 == 0: self.da.update_loss_("action4", 1) else: # ******* 截图 self.is_work = False self.one_static_time += 1 # 一次静止时间 if self.one_static_time % 20 == 0: self.da.update_loss_("action3", 1) # ******** self.action = ThreadedTCPRequestHandler.action # 键盘操作 if self.action is not None: # 往面板上写当前由于什么原因导致机器静止 if self.pre_action is None: pass if self.action_video is not None: if self.pre_action_video is None: pass video_recog_left() self.pre_action = self.action self.pre_action_video = self.action_video def data_read(self): pass def displayMessage(self, message): self.ui.textBrowser.append( '[' + time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())) + '] ' + message)
l2_reg = float(l2_reg) l1_reg = cli_args.get('l1_reg') if l1_reg != None: l1_reg = float(l1_reg) print(f'Running job with arguments\n{cli_args}') # define configs train_perf_filename = 'train-results.csv' test_perf_filename = 'test-results.csv' n_estimators = 5 if debug else 100 print(f'n_estimators: {n_estimators}') print(f'max_depth: {max_depth}') # init timer timer = Timer() # iterate over runs for run in range(runs): print(f'Starting run {run}') # load data data = load_sampled_data(sample_size) print(f'Loaded data with shape {data.shape}') # drop columns, onehot encode, or lookkup embeddings x, y = get_embedded_data(data, embedding_type, embedding_path, drop_columns) del data print(f'Encoded data shape: {x.shape}')
# model config window_size = int(cli_args.get('window_size', 5)) min_seq_length = int(cli_args.get('min_seq_length', 2)) embedding_size = int(cli_args.get('embedding_size', 300)) iters = int(cli_args.get('iters', 5)) desc = f'e{embedding_size}-w{window_size}-i{iters}-t{ts}' # I/O data_dir = os.environ['CMS_RAW'] curr_dir = os.path.join(proj_dir, 'cbow') embeddings_output = os.path.join(proj_dir, 'embeddings', f'cbow-{desc}.kv') loss_output = os.path.join(curr_dir, 'logs', f'train-loss-{desc}.csv') time_output = os.path.join(curr_dir, 'logs', f'train-time-{desc}.csv') # load corpus timer = Timer() corpus = load_hcpcs_corpus(debug) print(f'Loaded corpus with length {len(corpus)} in {timer.lap()}') # use sample for debug if debug: corpus = corpus[:500000] print(f'Using sample of corpus with length {len(corpus)}') # vocab size vocab_size = get_vocab_size(corpus) # loss and timing callback callback = GensimEpochCallback(loss_output, time_output) # train model
def main(experiment_name, optimizer, output_directory_root="experiments/resnet18_logistic_cifar10", epochs=60, batch_size=512, num_workers=1): output_directory = os.path.join(output_directory_root, experiment_name) if not os.path.isdir(output_directory): os.makedirs(output_directory, exist_ok=True) # Setup regular log file + tensorboard logfile_path = os.path.join(output_directory, "logfile.txt") setup_logger_tqdm(logfile_path) tensorboard_log_directory = os.path.join("runs", "resnet18_logistic_cifar10", experiment_name) tensorboard_summary_writer = SummaryWriter( log_dir=tensorboard_log_directory) # Choose Training Device use_cuda = torch.cuda.is_available() logger.info(f"CUDA Available? {use_cuda}") device = "cuda" if use_cuda else "cpu" # Datasets and Loaders train_set_loader, test_set_loader = get_data_loaders( batch_size, num_workers) # Create Model & Optimizer model = torchvision.models.resnet18(pretrained=True) for param in model.parameters(): param.requires_grad = False num_classes = 10 model.fc = nn.Linear(model.fc.in_features, 10) model.to(device) optimizer = optimizer(model.parameters()) logger.info("=========== Commencing Training ===========") logger.info(f"Epoch Count: {epochs}") logger.info(f"Batch Size: {batch_size}") # Load Checkpoint checkpoint_file_path = os.path.join(output_directory, "checkpoint.pth") start_epoch = 0 if os.path.exists(checkpoint_file_path): logger.info("Checkpoint Found - Loading!") checkpoint = torch.load(checkpoint_file_path) logger.info(f"Last completed epoch: {checkpoint['epoch']}") logger.info(f"Average Train Loss: {checkpoint['train_loss']}") logger.info(f"Top-1 Train Accuracy: {checkpoint['train_accuracy']}") logger.info(f"Top-1 Test Accuracy: {checkpoint['test_accuracy']}") start_epoch = checkpoint["epoch"] + 1 logger.info(f"Resuming at epoch {start_epoch}") model.load_state_dict(checkpoint["model_state_dict"]) optimizer.load_state_dict(checkpoint["optimizer_state_dict"]) else: logger.info("No checkpoint found, starting from scratch.") # Training Loop t = Timer() for epoch in range(start_epoch, epochs): t.start() logger.info("-" * 10) logger.info(f"Epoch {epoch}") logger.info("-" * 10) train_loss, train_accuracy = train_model(device, model, train_set_loader, optimizer) tensorboard_summary_writer.add_scalar("train_loss", train_loss, epoch) tensorboard_summary_writer.add_scalar("train_accuracy", train_accuracy, epoch) test_accuracy = test_model(device, model, test_set_loader, optimizer) tensorboard_summary_writer.add_scalar("test_accuracy", test_accuracy, epoch) # Save Checkpoint logger.info("Saving checkpoint.") torch.save( { 'epoch': epoch, 'model_state_dict': model.state_dict(), 'optimizer_state_dict': optimizer.state_dict(), 'train_loss': train_loss, 'train_accuracy': train_accuracy, 'test_accuracy': test_accuracy }, checkpoint_file_path) elapsed_time = t.stop() logger.info(f"End of epoch {epoch}, took {elapsed_time:0.4f} seconds.") logger.info(f"Average Train Loss: {train_loss}") logger.info(f"Top-1 Train Accuracy: {train_accuracy}") logger.info(f"Top-1 Test Accuracy: {test_accuracy}") logger.info("")
class XioAll(QtGui.QWidget): '''这个类为主程序类 ''' HOST = 'localhost' PORT = 8081 TOTAL = 0 isStatic = True action = None pre_action = None action_video = None # 视频内能识别 pre_action_video = None def __init__(self): super(XioAll, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.frame_left = None self.frame_right = None self.is_work = True self.one_static_time = 0 # 一次故障静止的时间 self.all_time = 0 # 一天的工作时间 self.q = MyQueue() # 存放帧队列,改为存放状态比较好 self.vision = Vision() # 若日期发生改变,自行插入全零数据 da = data_access.EquipmentTimeData() # 对损失项统计表进行操作 result_loss = da.select_("select * from loss ORDER BY SJ DESC limit 1") current_time = datetime.datetime.now().strftime('%Y-%m-%d') if str(result_loss[0][0]) != current_time: da.update('insert into loss(SJ,action1,action2,action3,action4,action5,action6)values' '("%s",%d,%d,%d,%d,%d,%d)' % (current_time, 0, 0, 0, 0, 0, 0)) else: pass da_oee = data_access.OEEData() # 对oee实时利用率进行统计 result_oee = da_oee.select_('select * from oee_date ORDER BY SJC DESC limit 1') if str(result_oee[0][0]) != current_time: da_oee.update_('insert into oee_date(SJC,O8,O9,O10,O11,O12,O13,O14,O15,O16,O17,O18)values' '("' + current_time + '",0,0,0,0,0,0,0,0,0,0,0)') else: pass self.thread_figure = Timer('updatePlay()', sleep_time=120) # 该线程用来每隔2分钟刷新绘图区 self.connect(self.thread_figure, QtCore.SIGNAL('updatePlay()'), self.draw) self.thread_figure.start() self.server = ThreadedTCPServer((self.HOST, self.PORT), ThreadedTCPRequestHandler) # 该线程用来一直监听客户端的请求 self.server_thread = threading.Thread(target=self.server.serve_forever) self.server_thread.start() self.thread_video_receive = threading.Thread(target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() def video_receive_local(self, cam1='./videos/left_cam.mp4', cam2='./videos/right_cam.mp4', time_flag=True): '''该方法用来接收本地视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :param time_flag: 是否休眠,本地视频为True :return: None ''' self.left_cam = cv2.VideoCapture(cam1) self.right_cam = cv2.VideoCapture(cam2) ret_1, frame_1 = self.left_cam.read() ret_2, frame_2 = self.right_cam.read() while True: self.frame_left = frame_1 self.frame_right = frame_2 if ret_1 is False: self.left_cam = cv2.VideoCapture(cam1) if ret_2 is False: self.right_cam = cv2.VideoCapture(cam2) ret_1, frame_1 = self.left_cam.read() ret_1, frame_2 = self.right_cam.read() if time_flag is True: time.sleep(0.04) def video_receive_rstp(self, cam1='rstp:', cam2='rstp:'): '''该方法用来接收网络视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :return: None ''' self.video_receive_local(cam1=cam1, cam2=cam2, time_flag=False) def video_play(self): '''该方法用来播放视频 :return: None ''' def label_show_left(frame, label=self.ui.label): # 左控件label播放 height, width, _ = frame.shape frame_change = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) frame_resize = cv2.resize(frame_change, (360, 240), interpolation=cv2.INTER_AREA) image = QtGui.QImage(frame_resize.data, frame_resize.shape[1], frame_resize.shape[0], QtGui.QImage.Format_RGB888) # 处理成QImage label.setPixmap(QtGui.QPixmap.fromImage(image)) def label_show_right(frame, label=self.ui.label_2): # 右空间Lable播放 label_show_left(frame, label) if self.frame_left is not None: label_show_left(self.frame_left) if self.frame_right is not None: label_show_right(self.frame_right) def draw(self): ''' 展示图标 :return: ''' def draw_fp(): # 绘制损失饼图 fp = Figure_Pie() da = data_access.EquipmentData() result = da.select() fp.plot(*(result[-1][1], result[-1][2], result[-1][3], result[-1][4])) # '*'有一个解包的功能,将(1,1,1,1)解包为 1 1 1 1 graphicscene_fp = QtGui.QGraphicsScene() graphicscene_fp.addWidget(fp.canvas) self.ui.graphicsView_Pie.setScene(graphicscene_fp) self.ui.graphicsView_Pie.show() def draw_oee(): # 绘制oee日推图 current_time = datetime.datetime.now().strftime('%Y-%m-%d') lossTime = data_access.EquipmentTimeData() result_loss = lossTime.select_("select * from loss ORDER BY SJ DESC limit 1") zongshijian = time.strftime('%H:%M:%S', time.localtime(time.time())) huanxing = result_loss[0][1] dailiao = result_loss[0][2] shebeiguzhang = result_loss[0][3] tingzhi = result_loss[0][4] # qitashijian=result[0][5] # kongyunzhuan=result[0][6] fuheshijian = (int(zongshijian.split(':')[0]) - 8) * 3600 + int(zongshijian.split(':')[1]) * 60 + int( zongshijian.split(':')[2]) - tingzhi shijijiagong_1 = fuheshijian - huanxing - dailiao - shebeiguzhang eff = int(shijijiagong_1 / fuheshijian * 100) # 计算效率 print(eff) hour = time.localtime()[3] # 实时更新 da_oee = data_access.OEEData() da_oee.update_("update oee_date set O" + str(hour) + "=" + str(eff) + ' where SJC="' + current_time + '"') L_eff = [] oee = Figure_OEE() da = data_access.OEEData() result = da.select() hour = time.localtime()[3] if hour < 20: for i in range(1, hour - 6): L_eff.append(result[-1][i]) oee.plot(*tuple(L_eff)) # 参数 graphicscene_oee = QtGui.QGraphicsScene() graphicscene_oee.addWidget(oee.canvas) self.ui.graphicsView_OEE.setScene(graphicscene_oee) self.ui.graphicsView_OEE.show() def draw_loss(): # 绘制损失直方图 loss = Figure_Loss() da = data_access.EquipmentTimeData() result = da.select() loss.plot(*(result[-1][1], result[-1][2], result[-1][3], result[-1][4])) graphicscene_loss = QtGui.QGraphicsScene() graphicscene_loss.addWidget(loss.canvas) self.ui.graphicsView_Loss.setScene(graphicscene_loss) self.ui.graphicsView_Loss.show() def draw_mt(): # 绘制耗材使用图 mt = Figure_MT() mt.plot() graphicscene_mt = QtGui.QGraphicsScene() graphicscene_mt.addWidget(mt.canvas) self.ui.graphicsView_MT.setScene(graphicscene_mt) self.ui.graphicsView_MT.show() draw_fp() draw_loss() draw_mt() draw_oee() def video_recog(self): ''' 视频识别部分 :return: ''' frame_left = self.frame_left # 原始彩色图,左边摄像头 frame_left_gray = cv2.cvtColor(frame_left, cv2.COLOR_BGR2GRAY) # 原始图的灰度图 # frame_right = self.frame_left # 原始彩色图 # frame_right_gray = cv2.cvtColor(frame_right, cv2.COLOR_BGR2GRAY) def video_recog_left(): img = frame_left spark = self.vision.find_spark(img) self.q.enqueue(spark) # print(spark) if spark or True in self.q.queue: # 如果一段间隔时间内不断有火花(和机器移动,稍后完成),则说明机器必定处于工作状态 #print('work') self.action_video = None self.one_static_time = 0 # 恢复到运动后,一次静止时间重新清零 else: # ******* 截图 self.one_static_time += 1 # 一次静止时间 if self.one_static_time % 60 == 0: print('start or static') print('静止了,往catch文件夹中查看原因') t = time.localtime() hour = t[3] mini = t[4] seco = t[5] filename = str(hour) + '-' + str(mini) + '-' + str(seco) cv2.imwrite('./catch/' + filename + '.jpg', img) # ******** self.action = ThreadedTCPRequestHandler.action # 键盘操作 if self.action is not None: # 往面板上写当前由于什么原因导致机器静止 if self.pre_action is None: print(self.action) message = '[' + time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())) + ']' + str(self.action) self.displayMessage(message) if self.vision.tiaoshi(frame_left_gray): self.action_video = 'tiaoshi' if self.action_video is not None: if self.pre_action_video is None: print(self.action_video) message = '[' + time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())) + ']' + str(self.action_video) self.displayMessage(message) def video_recog_right(): # 以后用来做换气瓶等的实现 pass video_recog_left() video_recog_right() self.pre_action = self.action self.pre_action_video = self.action_video def data_read(self): pass def displayMessage(self, message): self.ui.textBrowser.append(message)
def inference_mean_exemplar( model, current_epoch, current_iter, local_rank, data_loader, dataset_name, device="cuda", max_instance=3200, mute=False, ): model.train(False) # convert to a torch.device for efficiency device = torch.device(device) if not mute: logger = logging.getLogger("maskrcnn_benchmark.inference") logger.info("Start evaluation") total_timer = Timer() inference_timer = Timer() total_timer.tic() torch.cuda.empty_cache() if not mute: pbar = tqdm(total=len(data_loader), desc="Validation in progress") with torch.no_grad(): all_pred_obj, all_truth_obj, all_pred_attr, all_truth_attr = [], [], [], [] obj_loss_all, attr_loss_all = 0, 0 cnt = 0 for iteration, out_dict in enumerate(data_loader): if type(max_instance) is int: if iteration == max_instance // model.cfg.EXTERNAL.BATCH_SIZE: break if type(max_instance) is float: if iteration > max_instance * len( data_loader) // model.cfg.EXTERNAL.BATCH_SIZE: break # print(iteration) images = torch.stack(out_dict['images']) obj_labels = torch.cat(out_dict['object_labels'], -1) attr_labels = torch.cat(out_dict['attribute_labels'], -1) cropped_image = torch.stack(out_dict['cropped_image']) images = images.to(device) obj_labels = obj_labels.to(device) attr_labels = attr_labels.to(device) cropped_image = cropped_image.to(device) # loss_dict = model(images, targets) pred_obj = model.mean_of_exemplar_classify(cropped_image) all_pred_obj.extend(to_list(pred_obj)) all_truth_obj.extend(to_list(obj_labels)) cnt += 1 if not mute: pbar.update(1) obj_f1 = f1_score(all_truth_obj, all_pred_obj, average='micro') #attr_f1 = f1_score(all_truth_attr, all_pred_attr, average='micro') obj_loss_all /= (cnt + 1e-10) # wait for all processes to complete before measuring the time total_time = total_timer.toc() model.train(True) return obj_f1, 0, len(all_truth_obj)
class XioPlayVideo(QtGui.QWidget): '''这个类为主程序类 ''' def __init__(self): super(XioPlayVideo, self).__init__() self.ui = ui.Ui_Form() self.ui.setupUi(self) self.left_cam = cv2.VideoCapture('./videos/left_cam.mp4') # 左摄像头 self.right_cam = cv2.VideoCapture('./videos/right_cam.mp4') self.frame_left = None self.frame_right = None self.thread_video_receive = threading.Thread( target=self.video_receive_local) # 该线程用来读取视频流 self.thread_video_receive.start() self.thread_time = Timer('updatePlay()') # 该线程用来每隔0.04秒在label上绘图 self.connect(self.thread_time, QtCore.SIGNAL('updatePlay()'), self.video_play) self.thread_time.start() self.thread_recog = Timer('updatePlay()', sleep_time=1) # 该线程用来每隔一秒分析图像 self.connect(self.thread_recog, QtCore.SIGNAL('updatePlay()'), self.video_recog) self.thread_recog.start() self.thread_data = Timer('updatePlay()', sleep_time=1800) # 该线程用来每隔半小时向数据库读取数据 self.connect(self.thread_data, QtCore.SIGNAL('updatePlay()'), self.data_read) self.thread_data.start() self.thread_tcp = None # 该线程用来完成tcp,未写完 def video_receive_local(self, cam1='./videos/left_cam.mp4', cam2='./videos/right_cam.mp4', time_flag=True): '''该方法用来接收本地视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :param time_flag: 是否休眠,本地视频为True :return: None ''' if self.left_cam.isOpened() is False: self.left_cam = cv2.VideoCapture(cam1) if self.right_cam.isOpened() is False: self.right_cam = cv2.VideoCapture(cam2) ret_1, frame_1 = self.left_cam.read() ret_2, frame_2 = self.right_cam.read() while True: self.frame_left = frame_1 self.frame_right = frame_2 if ret_1 is False: self.left_cam = cv2.VideoCapture(cam1) if ret_2 is False: self.right_cam = cv2.VideoCapture(cam2) ret_1, frame_1 = self.left_cam.read() ret_1, frame_2 = self.right_cam.read() if time_flag is True: time.sleep(0.04) def video_receive_rstp(self, cam1='rstp:', cam2='rstp:'): '''该方法用来接收网络视频 :param cam1: 左摄像头数据源 :param cam2: 右摄像头数据源 :return: None ''' self.video_receive_local(cam1=cam1, cam2=cam2, time_flag=False) def video_play(self): '''该方法用来播放视频 :return: None ''' def label_show_left(frame, label=self.ui.label): # 左控件label播放 height, width, _ = frame.shape frame_change = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) frame_resize = cv2.resize(frame_change, (500, 300), interpolation=cv2.INTER_AREA) image = QtGui.QImage(frame_resize.data, frame_resize.shape[1], frame_resize.shape[0], QtGui.QImage.Format_RGB888) # 处理成QImage label.setPixmap(QtGui.QPixmap.fromImage(image)) def label_show_right(frame, label=self.ui.label_2): # 右空间Lable播放 label_show_left(frame, label) if self.frame_left is not None: label_show_left(self.frame_left) if self.frame_right is not None: label_show_right(self.frame_right) def video_recog(self): pass def data_read(self): pass
def pretext_train(args, recorder): if args.gpus is not None: print("Use GPU: {} for pretext training".format(args.gpus)) num_class, data_length, image_tmpl = pt_data_config(args) # print("tp_length is: ", data_length) train_transforms, test_transforms, eval_transforms = pt_augmentation_config( args) train_loader, val_loader, eval_loader, train_samples, val_samples, eval_samples = pt_data_loader_init( args, data_length, image_tmpl, train_transforms, test_transforms, eval_transforms) n_data = len(train_loader) model, model_ema = pt_model_config(args, num_class) # == optim config== contrast, criterion, optimizer = pt_optim_init(args, model, n_data) model = model.cuda() # == load weights == model, model_ema = pt_load_weight(args, model, model_ema, optimizer, contrast) if args.pt_method in ['dsm', 'moco']: model_ema = model_ema.cuda() # copy weights from `model' to `model_ema' moment_update(model, model_ema, 0) cudnn.benchmark = True # optionally resume from a checkpoint args.start_epoch = 1 # ==================================== our data augmentation method================================= if args.pt_method in ['dsm', 'dsm_triplet']: pos_aug = GenPositive() neg_aug = GenNegative() # =======================================add message ===================== recorder.record_message('a', '=' * 100) recorder.record_message('a', '-' * 40 + 'pretrain' + '-' * 40) recorder.record_message('a', '=' * 100) # ====================update lr_decay from str to numpy========= iterations = args.pt_lr_decay_epochs.split(',') args.pt_lr_decay_epochs = list([]) for it in iterations: args.pt_lr_decay_epochs.append(int(it)) timer = Timer() # routine print('*' * 70 + 'Step1: pretrain' + '*' * 20 + '*' * 50) for epoch in range(args.pt_start_epoch, args.pt_epochs + 1): timer.tic() pt_adjust_learning_rate(epoch, args, optimizer) print("==> training...") time1 = time.time() if args.pt_method == "moco": loss, prob = train_moco(epoch, train_loader, model, model_ema, contrast, criterion, optimizer, args, recorder) elif args.pt_method == "dsm": loss, prob = train_dsm(epoch, train_loader, model, model_ema, contrast, criterion, optimizer, args, pos_aug, neg_aug, recorder) # loss, prob = epoch * 0.01, 0.02*epoch elif args.pt_method == "dsm_triplet": loss = train_dsm_triplet(epoch, train_loader, model, optimizer, args, pos_aug, neg_aug, recorder) else: Exception("Not support method now!") recorder.record_pt_train(loss) time2 = time.time() print('epoch {}, total time {:.2f}'.format(epoch, time2 - time1)) timer.toc() left_time = timer.average_time * (args.pt_epochs - epoch) message = "Step1: pretrain now loss is: {} left time is : {} now is: {}".format( loss, timer.format(left_time), datetime.now()) print(message) recorder.record_message('a', message) state = { 'opt': args, 'model': model.state_dict(), 'contrast': contrast.state_dict(), 'optimizer': optimizer.state_dict(), 'epoch': epoch, } recorder.save_pt_model(args, state, epoch) print("finished pretrain, the trained model is record in: {}".format( recorder.pt_checkpoint)) return recorder.pt_checkpoint