def initialize_model(path_classes, path_anchors, path_model, input_shape,
freeze_body, path_weights, td_len, mode, spp, loss_percs,
**kwargs):
class_names = ktrain.get_classes(path_classes)
num_classes = len(class_names)
anchors = ktrain.get_anchors(path_anchors)
# Create model
title = 'Create Keras model'
print('{} {} {}'.format('=' * print_indnt, title,
'=' * (print_line - 2 - len(title) - print_indnt)))
model = create_model(input_shape,
anchors,
num_classes,
freeze_body=freeze_body,
weights_path=path_weights,
td_len=td_len,
mode=mode,
spp=spp,
loss_percs=loss_percs)
# Store model architecture
model_architecture = model.to_json()
with open(path_model + 'architecture.json', 'w') as f:
json.dump(model_architecture, f)
print("Model architecture stored as json")
print('=' * print_line)
# Training callbacks
callbacks = {
'logging':
TensorBoard(log_dir=path_model),
'checkpoint':
ModelCheckpoint(
path_model + 'weights/' +
'ep{epoch:03d}-loss{loss:.5f}-val_loss{val_loss:.5f}.h5',
monitor='val_loss',
save_weights_only=True,
save_best_only=True,
period=1),
'reduce_lr_1':
ReduceLROnPlateau(monitor='loss',
min_delta=0.5,
factor=0.1,
patience=4,
verbose=1),
'reduce_lr_2':
ReduceLROnPlateau(monitor='val_loss',
min_delta=0,
factor=0.1,
patience=4,
verbose=1),
'early_stopping':
EarlyStopping(monitor='val_loss', min_delta=0, patience=10, verbose=1)
}
log('MODEL CREATED')
return model, callbacks, anchors, num_classes
def main(path_results, dataset_name, model_num, score, iou, num_annotation_file=1,
plot=True, full=True, best_weights=True):
model_folder = train_utils.get_model_path(path_results, dataset_name, model_num)
train_params = json.load(open(model_folder + 'train_params.json', 'r'))
class_names = ktrain.get_classes(train_params['path_classes'])
# train_params['path_annotations'][1] = train_params['path_annotations'][1].replace('_sk10', '_sk20')
annotations_file = train_params['path_annotations'][num_annotation_file]
if 'adl' in annotations_file and train_params.get('size_suffix', '') != '':
annotations_file = annotations_file.replace(train_params.get('size_suffix', ''), '')
# annotations_file = annotations_file.replace('.txt', '_pr416.txt')
print(' * Exploring:', annotations_file)
if best_weights:
preds_filename = '{}preds_{}_score{}_iou{}.json'.format(model_folder, annotations_file.split('/')[-1][:-4], score, iou)
eval_filename = '{}stats_{}_score{}_iou{}.json'.format(model_folder, annotations_file.split('/')[-1][:-4], score, iou)
else:
preds_filename = '{}preds_stage2_{}_score{}_iou{}.json'.format(model_folder, annotations_file.split('/')[-1][:-4], score, iou)
eval_filename = '{}stats_stage2_{}_score{}_iou{}.json'.format(model_folder, annotations_file.split('/')[-1][:-4], score, iou)
# print(preds_filename)
print('='*80)
train_diff, train_loss, val_loss = get_train_resume(model_folder)
_ = prediction_utils.predict_and_store_from_annotations(model_folder, train_params, annotations_file, preds_filename, score, iou, best_weights=best_weights)
occurrences = None
eval_stats, videos = get_full_evaluation(annotations_file, preds_filename, class_names, full)
# resume = get_excel_resume(model_folder, train_params, train_loss, val_loss, eval_stats, train_diff, fps, score, iou)
plot_prediction_resume(eval_stats, videos, class_names, 'video', annotations_file, model_num, plot);
occurrences = plot_prediction_resume(eval_stats, videos, class_names, 'class', annotations_file, model_num, plot)
# print(resume)
return model_folder, class_names, videos, occurrences, eval_stats, train_params, (train_loss, val_loss, train_diff)
print('='*80)
print(preds_num[label][preds_num[label]['image_id'] == image_id][['bbox', 'score']])
boxes = preds_num[label][preds_num[label]['image_id'] == image_id]
# boxes = boxes.apply(lambda r: ','.join([ str(b) for b in r['bbox'] ]) + ',' + str(r['category_id']), axis=1).tolist()
sample = '{}{}.jpg'.format(path_dataset, image_id)
for i,r in boxes.iterrows():
bbox = r['bbox']
sample += ' {},{},{},{},{}'.format(bbox[0], bbox[1], bbox[0]+bbox[2], bbox[1]+bbox[3], r['category_id'])
result = print_annotations(sample, class_names, colors, perc=0.4)
# result = print_annotations(sample, perc=0.4, class_names=class_names)
results.append(result)
return results
class_names = ktrain.get_classes(train_params['path_classes'])
hsv_tuples = [(x / len(class_names), 1., 1.)
for x in range(len(class_names))]
colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples))
colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), colors))
preds_count = preds_count[preds_count['320'] != 0]
#image_id = preds_count.index[1]
for image_id in preds_count.index:
print('='*80)
print('='*80)
print(image_id)
def main():
import os
os.chdir(os.getcwd() + '/..')
# %%
import train_utils
import pandas as pd
import sys
sys.path.append('keras_yolo3/')
sys.path.append('keras_yolo3/yolo3/')
import keras_yolo3.train as ktrain
##path_dataset = '/mnt/hdd/datasets/VOC/'
#path_results = '/mnt/hdd/egocentric_results/'
##path_classes = './dataset_scripts/coco/coco_classes.txt'
#path_dataset = '/mnt/hdd/datasets/coco/'
#path_classes = './dataset_scripts/adl/adl_classes_v3_8.txt'
#
##gt_filename = '/home/asabater/projects/Egocentric-object-detection/dataset_scripts/coco/annotations_coco_val_coco.json'
##gt = json.load(open(gt_filename, 'r'))['annotations']
#
#score = 0.4
#gt_filename = path_results + 'default/voc_yolo_model_0/preds_annotations_coco_val_score{}_iou.json'.format(score)
##gt_filename = path_results + 'adl/0414_2252_model_20/preds_annotations_adl_val_v3_8_r_fd10_fsn1_score0_iou0.5.json'
##gt_filename = path_results + 'adl/0414_2252_model_20/preds_annotations_adl_train_v3_8_r_fd10_fsn1_score5e-05_iou0.5.json'
#gt = json.load(open(gt_filename, 'r'))
#path_classes = './dataset_scripts/voc/voc_classes.txt'
#path_dataset = '/mnt/hdd/datasets/VOC/'
#score = 0.4
#gt_filename = path_results + 'voc/0326_1706_model_0/preds_annotations_voc_val_score{}_iou.json'.format(score)
#gt = json.load(open(gt_filename, 'r'))
path_results = '/mnt/hdd/egocentric_results/'
dataset_name = 'adl'
path_dataset = '/home/asabater/projects/ADL_dataset/'
model_num = 62
model_folder = train_utils.get_model_path(path_results, dataset_name,
model_num)
train_params = json.load(open(model_folder + 'train_params.json', 'r'))
annotations_file = train_params['path_annotations'][1]
preds_filename = '{}preds_stage2_{}_score{}_iou{}.json'.format(
model_folder,
annotations_file.split('/')[-1][:-4], 0, 0.5)
gt = json.load(open(preds_filename, 'r'))
# %%
preds = pd.DataFrame([p for p in gt if p['score'] >= 0.3])
preds = [{
'image_id': i,
'bboxes': g.bbox.tolist(),
'category_id': g.category_id.tolist()
} for i, g in preds.groupby('image_id')]
# %%
from playground.test_annotations import print_annotations
class_names = ktrain.get_classes(train_params['path_classes'])
hsv_tuples = [(x / len(class_names), 1., 1.)
for x in range(len(class_names))]
colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples))
colors = list(
map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)),
colors))
preds = preds[random.randint(0, len(preds)):]
for pred in preds:
sample = '{}{}.jpg'.format(path_dataset, pred['image_id'])
for bbox, cat in zip(pred['bboxes'], pred['category_id']):
sample += ' {},{},{},{},{}'.format(bbox[0], bbox[1],
bbox[0] + bbox[2],
bbox[1] + bbox[3], cat)
print(sample)
pass
# result = print_annotations(sample, 0.7, class_names)
result = print_annotations(sample, class_names, colors, perc=0.7)
cv2.imshow("result", result)
if cv2.waitKey(200) & 0xFF == ord('q'):
cv2.destroyAllWindows()
break
from dataset_scripts.annotations_to_coco import annotations_to_coco
from dataset_scripts.generate_custom_anchors import EYOLO_Kmeans
remove_empty_frames = True
cluster_number = 9
dataset_version = ''
path_annotations = [
'./dataset_scripts/kitchen/annotations_kitchen_train{}.txt'.format(
dataset_version),
'./dataset_scripts/kitchen/annotations_kitchen_val{}.txt'.format(
dataset_version)
]
path_classes = './dataset_scripts/kitchen/kitchen_classes{}.txt'.format(
dataset_version)
class_names = ktrain.get_classes(path_classes)
merge_classes = [
([
'spoon', 'knife', 'fork', 'cutlery', 'ladle', 'utensil', 'spatula',
'tongs'
], 'cutlery'),
(['tap'], 'tap'),
(['plate', 'bowl'], 'plate/bowl'),
(['fridge', 'freezer'], 'fridge/freezer'),
(['pan', 'pot', 'cooker:slow'], 'pan/pot'),
(['salt', 'oil', 'vinegar'], 'salt/oil/vinegar'),
(['cup', 'glass'], 'cup/glass'),
([
'bottle', 'jar', 'jug', 'liquid', 'liquid:washing', 'drink', 'beer',
'coke', 'can', 'soap'
def main_v2(path_results, dataset_name, model_num, input_shape=None, eval_train=True, full=False):
model_folder = train_utils.get_model_path(path_results, dataset_name, model_num)
train_params = json.load(open(model_folder + 'train_params.json', 'r'))
class_names = ktrain.get_classes(train_params['path_classes'])
annotation_files = train_params['path_annotations']
# annotation_files[1] = annotation_files[1].replace('_sk20', '_sk3')
# train_params['eval_val_score'] = 0.05
if input_shape is not None:
train_params['input_shape'] = input_shape
# train_params['eval_val_score'] = 0.005
# train_params['eval_train_score'] = 0.005
# Pred and eval val bw
_, preds_filename_val_bw = prediction_utils.predict_and_store_from_annotations(model_folder,
train_params, annotation_files[1], model_folder,
image_size=train_params['input_shape'],
score=train_params.get('eval_val_score'), nms_iou=0.5,
best_weights=True, raw_eval='def')
eval_stats_val_bw, videos = get_full_evaluation(annotation_files[1], preds_filename_val_bw, class_names, full)
# Pred and eval val st2
_, preds_filename_val_st2 = prediction_utils.predict_and_store_from_annotations(model_folder,
train_params, annotation_files[1], model_folder,
train_params['input_shape'],
score=train_params.get('eval_val_score'), nms_iou=0.5,
best_weights=False, raw_eval='def')
eval_stats_val_st2, videos = get_full_evaluation(annotation_files[1], preds_filename_val_st2, class_names, full)
print('='*80)
print('mAP@50 | val_bw: {:.2f} | val_st2: {:2f}'.format(eval_stats_val_bw['total'][1]*100, eval_stats_val_st2['total'][1]*100))
print('='*80)
# Pred and eval train
best_weights = eval_stats_val_bw['total'][1] > eval_stats_val_st2['total'][1]
if eval_train:
_, preds_filename_train = prediction_utils.predict_and_store_from_annotations(model_folder,
train_params, annotation_files[0], model_folder,
train_params['input_shape'],
score=train_params.get('eval_train_score'), nms_iou=0.5,
best_weights=best_weights, raw_eval='def')
eval_stats_train, videos = get_full_evaluation(annotation_files[0], preds_filename_train, class_names, full)
else:
eval_stats_train = None
train_diff, train_loss, val_loss = get_train_resume(model_folder)
eval_stats_val = eval_stats_val_bw if best_weights else eval_stats_val_st2
full_resume = get_excel_resume_full(model_folder, train_params,
train_loss, val_loss,
eval_stats_train, eval_stats_val, train_diff,
best_weights)
print('='*80)
print(full_resume)
pyperclip.copy(full_resume)
print('='*80)
if eval_train:
print('mAP@50 | val_bw: {:.4f} | val_st2: {:.4f} | train: {:.4f}'.format(
eval_stats_val_bw['total'][1]*100, eval_stats_val_st2['total'][1]*100, eval_stats_train['total'][1]*100
))
print('R10 | val_bw: {:.4f} | val_st2: {:.4f} | train: {:.4f}'.format(
eval_stats_val_bw['total'][7]*100, eval_stats_val_st2['total'][7]*100, eval_stats_train['total'][7]*100
))
else:
print('mAP@50 | val_bw: {:.4f} | val_st2: {:.4f}'.format(
eval_stats_val_bw['total'][1]*100, eval_stats_val_st2['total'][1]*100
))
print('R10 | val_bw: {:.4f} | val_st2: {:.4f}'.format(
eval_stats_val_bw['total'][7]*100, eval_stats_val_st2['total'][7]*100
))
print('='*80)
# return full_resume
return eval_stats_train, eval_stats_val, videos, class_names