archive/runapp.py [338:559]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - data = bottle.request.json images_to_label = [im['id'] for im in data['images']] label_to_assign = data['label'] # Use image ids in images_to_label to get the corresponding dataset indices indices_to_label = [] indices_detection_ids = [dataset.samples[i][0] for i in indices] for im in images_to_label: pos = indices_detection_ids.index(im) ind = indices[pos] indices_to_label.append(ind) label_category_name = label_to_assign.lower() if label_category_name == 'empty': # Update records in dataset dataloader but not in the PostgreSQL database moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.UserDetection.value, indices_to_label) # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) else: # Get the category id for the assigned label existing_category_entries = {cat.name: cat.id for cat in Category.select()} try: label_category_id = existing_category_entries[label_category_name] except: print('The label was not found in the database Category table') raise NotImplementedError # Update entries in the PostgreSQL database ## get Detection table entries corresponding to the images being labeled matching_detection_entries = (Detection .select(Detection.id, Detection.category_id) .where((Detection.id << images_to_label))) # << means IN ## update category_id, category_confidence, and kind of each Detection entry in the PostgreSQL database for mde in matching_detection_entries: command = Detection.update(category_id=label_category_id, category_confidence=1, kind=DetectionKind.UserDetection.value).where(Detection.id == mde.id) command.execute() # Update records in dataset dataloader for il in indices_to_label: sample_data = list(dataset.samples[il]) sample_data[1] = label_category_id sample_data[2] = DetectionKind.UserDetection.value sample_data[3] = 1 dataset.samples[il] = tuple(sample_data) moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.UserDetection.value, indices_to_label) # print(set(dataset.set_indices[4]).update(set(indices_to_label))) dataset.set_indices[4] = list(set(dataset.set_indices[4]).union(set(indices_to_label))) # add the index to the set of labeled/confirmed indices # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) print([len(x) for x in dataset.set_indices]) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/confirmPredictedLabel', method='POST') def confirm_predicted_label(): global dataset global indices data = bottle.request.json image_to_label = data['image'] label_to_assign = data['label'] # Use image id images_to_label to get the corresponding dataset index indices_detection_ids = [dataset.samples[i][0] for i in indices] pos = indices_detection_ids.index(image_to_label) index_to_label = indices[pos] label_category_name = label_to_assign.lower() if label_category_name == 'empty': # Update records in dataset dataloader but not in the PostgreSQL database moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.ConfirmedDetection.value, [index_to_label]) # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) # userdetection + confirmed detection? else: # Get the category id for the assigned label existing_category_entries = {cat.name: cat.id for cat in Category.select()} try: label_category_id = existing_category_entries[label_category_name] except: print('The label was not found in the database Category table') raise NotImplementedError # Update entries in the PostgreSQL database ## get Detection table entries corresponding to the images being labeled matching_detection_entries = (Detection .select(Detection.id, Detection.category_id) .where((Detection.id==image_to_label))) # << means IN ## update category_id, category_confidence, and kind of each Detection entry in the PostgreSQL database mde = matching_detection_entries.get() command = Detection.update(category_id=label_category_id, category_confidence=1, kind=DetectionKind.ConfirmedDetection.value).where(Detection.id == mde.id) command.execute() # Update records in dataset dataloader sample_data = list(dataset.samples[index_to_label]) sample_data[1] = label_category_id sample_data[2] = DetectionKind.ConfirmedDetection.value sample_data[3] = 1 dataset.samples[index_to_label] = tuple(sample_data) moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.ConfirmedDetection.value, [index_to_label]) dataset.set_indices[4] = list(set(dataset.set_indices[4]).union({index_to_label})) # add the index to the set of labeled/confirmed indices # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) print([len(x) for x in dataset.set_indices]) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/trainClassifier', method='POST') def train_classifier(): global dataset global kwargs global sampler global classifier_trained global X_pred global y_pred data = bottle.request.json # Train on samples that have been labeled so far # dataset.set_kind(DetectionKind.UserDetection.value) dataset.set_kind(4) print(dataset.current_set) print(type(dataset.current_set)) X_train = dataset.em[dataset.current_set] y_train = np.asarray(dataset.getlabels()) # print(y_train) timer = time.time() kwargs["model"].fit(X_train, y_train) print('Training took %0.2f seconds'%(time.time() - timer)) timer = time.time() joblib.dump(kwargs["model"], "%s/%s_%04d.skmodel"%(args.checkpoint_dir, 'classifier', len(dataset.current_set))) print('Saving classifier checkpoint took %0.2f seconds'%(time.time() - timer)) # Predict on the samples that have not been labeled timer = time.time() dataset.set_kind(DetectionKind.ModelDetection.value) X_pred = dataset.em[dataset.current_set] y_pred = kwargs["model"].predict(X_pred) print('Predicting on unlabeled samples took %0.2f seconds'%(time.time() - timer)) # print(y_pred) # Update model predicted class in PostgreSQL database # timer = time.time() # for pos in range(len(y_pred)): # idx = dataset.current_set[pos] # det_id = dataset.samples[idx][0] # matching_detection_entries = (Detection # .select(Detection.id, Detection.category_id) # .where((Detection.id == det_id))) # mde = matching_detection_entries.get() # command = Detection.update(category_id=y_pred[pos]).where(Detection.id == mde.id) # command.execute() # print('Updating the database took %0.2f seconds'%(time.time() - timer)) # Alternative: batch update PostgreSQL database # timer = time.time() # det_ids = [dataset.samples[dataset.current_set[pos]][0] for pos in range(len(y_pred))] # y_pred = [int(y) for y in y_pred] # det_id_pred_pairs = list(zip(det_ids, y_pred)) # case_statement = Case(Detection.id, det_id_pred_pairs) # command = Detection.update(category_id=case_statement).where(Detection.id.in_(det_ids)) # command.execute() # print('Updating the database the other way took %0.2f seconds'%(time.time() - timer)) # Update dataset dataloader timer = time.time() for pos in range(len(y_pred)): idx = dataset.current_set[pos] sample_data = list(dataset.samples[idx]) sample_data[1] = y_pred[pos] dataset.samples[idx] = tuple(sample_data) print('Updating the dataset dataloader took %0.2f seconds'%(time.time() - timer)) if not classifier_trained: # once the classifier has been trained the first time, switch to AL sampling classifier_trained = True sampler = get_AL_sampler('confidence')(dataset.em, dataset.getalllabels(), 1234) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/showFullsizeImage', method='POST') def show_fullsize_image(): data = bottle.request.json image_src = data['img_src'] matching_image_entries = (Image .select(Image.file_name, Image.source_file_name) .where((Image.file_name == image_src))) try: mie = matching_image_entries.get() data['success_status'] = True data['fullsize_src'] = mie.source_file_name except: data['success_status'] = False bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/getSequentialImages', method='POST') def get_sequential_images(): data = bottle.request.json image_src = data['img_src'] matching_image_entries = (Image .select(Image.seq_id, Image.seq_num_frames, Image.frame_num) .where((Image.file_name == image_src))) try: mie = matching_image_entries.get() if mie.seq_num_frames > 1: images_in_seq = (Image .select(Image.source_file_name) .where((Image.seq_id == mie.seq_id)) .order_by(Image.frame_num)) image_sequence = sorted(list(set([i.source_file_name for i in images_in_seq]))) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - research/active_learning/labeling_tool/runapp.py [340:561]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - data = bottle.request.json images_to_label = [im['id'] for im in data['images']] label_to_assign = data['label'] # Use image ids in images_to_label to get the corresponding dataset indices indices_to_label = [] indices_detection_ids = [dataset.samples[i][0] for i in indices] for im in images_to_label: pos = indices_detection_ids.index(im) ind = indices[pos] indices_to_label.append(ind) label_category_name = label_to_assign.lower() if label_category_name == 'empty': # Update records in dataset dataloader but not in the PostgreSQL database moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.UserDetection.value, indices_to_label) # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) else: # Get the category id for the assigned label existing_category_entries = {cat.name: cat.id for cat in Category.select()} try: label_category_id = existing_category_entries[label_category_name] except: print('The label was not found in the database Category table') raise NotImplementedError # Update entries in the PostgreSQL database ## get Detection table entries corresponding to the images being labeled matching_detection_entries = (Detection .select(Detection.id, Detection.category_id) .where((Detection.id << images_to_label))) # << means IN ## update category_id, category_confidence, and kind of each Detection entry in the PostgreSQL database for mde in matching_detection_entries: command = Detection.update(category_id=label_category_id, category_confidence=1, kind=DetectionKind.UserDetection.value).where(Detection.id == mde.id) command.execute() # Update records in dataset dataloader for il in indices_to_label: sample_data = list(dataset.samples[il]) sample_data[1] = label_category_id sample_data[2] = DetectionKind.UserDetection.value sample_data[3] = 1 dataset.samples[il] = tuple(sample_data) moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.UserDetection.value, indices_to_label) # print(set(dataset.set_indices[4]).update(set(indices_to_label))) dataset.set_indices[4] = list(set(dataset.set_indices[4]).union(set(indices_to_label))) # add the index to the set of labeled/confirmed indices # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) print([len(x) for x in dataset.set_indices]) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/confirmPredictedLabel', method='POST') def confirm_predicted_label(): global dataset global indices data = bottle.request.json image_to_label = data['image'] label_to_assign = data['label'] # Use image id images_to_label to get the corresponding dataset index indices_detection_ids = [dataset.samples[i][0] for i in indices] pos = indices_detection_ids.index(image_to_label) index_to_label = indices[pos] label_category_name = label_to_assign.lower() if label_category_name == 'empty': # Update records in dataset dataloader but not in the PostgreSQL database moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.ConfirmedDetection.value, [index_to_label]) # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) # userdetection + confirmed detection? else: # Get the category id for the assigned label existing_category_entries = {cat.name: cat.id for cat in Category.select()} try: label_category_id = existing_category_entries[label_category_name] except: print('The label was not found in the database Category table') raise NotImplementedError # Update entries in the PostgreSQL database ## get Detection table entries corresponding to the images being labeled matching_detection_entries = (Detection .select(Detection.id, Detection.category_id) .where((Detection.id==image_to_label))) # << means IN ## update category_id, category_confidence, and kind of each Detection entry in the PostgreSQL database mde = matching_detection_entries.get() command = Detection.update(category_id=label_category_id, category_confidence=1, kind=DetectionKind.ConfirmedDetection.value).where(Detection.id == mde.id) command.execute() # Update records in dataset dataloader sample_data = list(dataset.samples[index_to_label]) sample_data[1] = label_category_id sample_data[2] = DetectionKind.ConfirmedDetection.value sample_data[3] = 1 dataset.samples[index_to_label] = tuple(sample_data) moveRecords(dataset, DetectionKind.ModelDetection.value, DetectionKind.ConfirmedDetection.value, [index_to_label]) dataset.set_indices[4] = list(set(dataset.set_indices[4]).union({index_to_label})) # add the index to the set of labeled/confirmed indices # numLabeled = len(dataset.set_indices[DetectionKind.UserDetection.value]) print([len(x) for x in dataset.set_indices]) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/trainClassifier', method='POST') def train_classifier(): global dataset global kwargs global sampler global classifier_trained global X_pred global y_pred data = bottle.request.json # Train on samples that have been labeled so far # dataset.set_kind(DetectionKind.UserDetection.value) dataset.set_kind(4) print(dataset.current_set) print(type(dataset.current_set)) X_train = dataset.em[dataset.current_set] y_train = np.asarray(dataset.getlabels()) # print(y_train) timer = time.time() kwargs["model"].fit(X_train, y_train) print('Training took %0.2f seconds'%(time.time() - timer)) timer = time.time() joblib.dump(kwargs["model"], "%s/%s_%04d.skmodel"%(args.checkpoint_dir, 'classifier', len(dataset.current_set))) print('Saving classifier checkpoint took %0.2f seconds'%(time.time() - timer)) # Predict on the samples that have not been labeled timer = time.time() dataset.set_kind(DetectionKind.ModelDetection.value) X_pred = dataset.em[dataset.current_set] y_pred = kwargs["model"].predict(X_pred) print('Predicting on unlabeled samples took %0.2f seconds'%(time.time() - timer)) # print(y_pred) # Update model predicted class in PostgreSQL database # timer = time.time() # for pos in range(len(y_pred)): # idx = dataset.current_set[pos] # det_id = dataset.samples[idx][0] # matching_detection_entries = (Detection # .select(Detection.id, Detection.category_id) # .where((Detection.id == det_id))) # mde = matching_detection_entries.get() # command = Detection.update(category_id=y_pred[pos]).where(Detection.id == mde.id) # command.execute() # print('Updating the database took %0.2f seconds'%(time.time() - timer)) # Alternative: batch update PostgreSQL database # timer = time.time() # det_ids = [dataset.samples[dataset.current_set[pos]][0] for pos in range(len(y_pred))] # y_pred = [int(y) for y in y_pred] # det_id_pred_pairs = list(zip(det_ids, y_pred)) # case_statement = Case(Detection.id, det_id_pred_pairs) # command = Detection.update(category_id=case_statement).where(Detection.id.in_(det_ids)) # command.execute() # print('Updating the database the other way took %0.2f seconds'%(time.time() - timer)) # Update dataset dataloader timer = time.time() for pos in range(len(y_pred)): idx = dataset.current_set[pos] sample_data = list(dataset.samples[idx]) sample_data[1] = y_pred[pos] dataset.samples[idx] = tuple(sample_data) print('Updating the dataset dataloader took %0.2f seconds'%(time.time() - timer)) if not classifier_trained: # once the classifier has been trained the first time, switch to AL sampling classifier_trained = True sampler = get_AL_sampler('confidence')(dataset.em, dataset.getalllabels(), 1234) bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/showFullsizeImage', method='POST') def show_fullsize_image(): data = bottle.request.json image_src = data['img_src'] matching_image_entries = (Image .select(Image.file_name, Image.source_file_name) .where((Image.file_name == image_src))) try: mie = matching_image_entries.get() data['success_status'] = True data['fullsize_src'] = mie.source_file_name except: data['success_status'] = False bottle.response.content_type = 'application/json' bottle.response.status = 200 return json.dumps(data) @webUIapp.route('/getSequentialImages', method='POST') def get_sequential_images(): data = bottle.request.json image_src = data['img_src'] matching_image_entries = (Image .select(Image.seq_id, Image.seq_num_frames, Image.frame_num) .where((Image.file_name == image_src))) try: mie = matching_image_entries.get() if mie.seq_num_frames > 1: images_in_seq = (Image .select(Image.source_file_name) .where((Image.seq_id == mie.seq_id)) .order_by(Image.frame_num)) image_sequence = sorted(list(set([i.source_file_name for i in images_in_seq]))) - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -