Skip to content

GitLab

Commit 94ce7a09 authored by Junior Silva Souza's avatar Junior Silva Souza
Browse files

Adicionando funcionalidades para gerar uma pasta completa com relatórios

parent 8e1928cc
Hide whitespace changes
Inline Side-by-side
Showing with 111 additions and 54 deletions
implementing2det3/experimenter.py
View file @ 94ce7a09
......@@ -5,9 +5,9 @@ import torchmetrics
import cv2
import numpy as np
#from mmdet.apis import init_detector, inference_detector
#from torchmetrics.detection import MeanAveragePrecision
#from torchmetrics.functional.detection import intersection_over_union
from mmdet.apis import init_detector, inference_detector
from torchmetrics.detection import MeanAveragePrecision
from torchmetrics.functional.detection import intersection_over_union
class ExtractCoco(object):
......@@ -129,13 +129,29 @@ class Boxes(object):
class Testing(object):
def __init__(self, model,path,prefix, device='cuda:0'):
def __init__(self, model,path_json,results_p,prefix,technique='ssd',threshold=0.3, threshold_class=0.3, device='cuda:0'):
try:
self.model = model
assert os.path.exists(path), 'This file not exists!'
self.coco = ExtractCoco(path)
self.device = torch.device(device) if torch.cuda.is_available() else torch.device('cpu')
self.prefix = prefix
assert os.path.exists(path_json), 'This file not exists!'
self.coco = ExtractCoco(path_json)
self.device = torch.device(device) if torch.cuda.is_available() else torch.device('cpu')
self.prefix = prefix
self.threshold = threshold
self.threshold_class = threshold_class
self.metrics = MeanAveragePrecision(iou_type="bbox")
self.fold = os.path.basename(path_json).split('_test')[0]
results_path = os.path.join(results_p,'results')
local_base = os.path.join(results_path,technique)
local_imgs_results = os.path.join(local_base,self.fold)
for p in [results_path,local_base,local_imgs_results]:
if not os.path.exists(p):
os.mkdir(p)
self.path = local_imgs_results
self.estatistics_local = results_path
self.technique = technique
if self.model is not None:
self.model.eval()
self.model.to(self.device)
......@@ -143,66 +159,107 @@ class Testing(object):
except ValueError as error:
print(error)
def running(self,qt_show=10):
def running(self,path='.',classes=('pig',),qt_show=10):
try:
imgs = self.coco.extract_dic()
dic_labels = self.coco.dic_labels
dic_labels = {value:key for key,value in self.coco.dic_labels.items()}
lab_dic = {dic_labels[cl]:i for i,cl in enumerate(classes) }
fp_t = tp_t = total = 0
for image in imgs:
img_p, targets, size = image
img = cv2.imread(os.path.join(self.prefix,img_p))
#preds = inference_detector(self.model, img)
#box_preds = (preds['boxes']).clone()
#box_labels = (preds['labels']).clone()
#box_scores = (preds['scores']).clone()
#labels = []
#scores = []
#boxes = []
#b_gt = []
#for gt in targets['boxes']:
# b_gt.append( [[g.item() for g in gt],1,0] )
# gt = (gt.unsqueeze(0)).to(self.device)
# distances_dic = {(intersection_over_union(pred.unsqueeze(0),gt)).item():[pred,i] for i,pred in enumerate(box_preds)}
# distances = list(distances_dic.keys())
# value = np.max(distances)
# if value >= self.threshold:
# box_closed = distances_dic[value]
# labels.append((box_labels[box_closed[1]]).item())
# scores.append((box_scores[box_closed[1]]).item())
# boxes.append(box_closed[0])
# box_scores = [v for i,v in enumerate(box_scores) if box_closed[1] !=i]
# box_labels = [v for i,v in enumerate(box_labels) if box_closed[1] !=i]
# box_preds = [v for i,v in enumerate(box_preds) if box_closed[1] !=i]
#boxes = [bx.cpu().tolist() for bx in boxes]
#boxes_fp = [[bx.cpu().tolist(),box_labels[i].item(),round(box_scores[i].item(),3)] for i,bx in enumerate(box_preds) if box_scores[i].item() >= self.threshold_class]
#preds_closed ={
# 'boxes' : torch.tensor(boxes),
# 'scores': torch.tensor(scores),
# 'labels': torch.tensor(labels),
#}
print(targets['labels'])
targets['labels'] = torch.tensor([lab_dic[ value.item() ] for value in targets['labels']])
preds = inference_detector(self.model, img)
box_preds = preds.pred_instances.bboxes.clone()
box_labels = preds.pred_instances.labels.clone()
box_scores = preds.pred_instances.scores.clone()
labels = []
scores = []
boxes = []
b_gt = []
for i,gt in enumerate(targets['boxes']):
b_gt.append( [[g.item() for g in gt],(targets['labels'])[i].item(),0] )
gt = (gt.unsqueeze(0)).to(self.device)
distances_dic = {(intersection_over_union(pred.unsqueeze(0),gt)).item():[pred,i] for i,pred in enumerate(box_preds)}
values_iou = list(distances_dic.keys())
value = np.max(values_iou)
if value >= self.threshold:
box_closed = distances_dic[value]
#print('Target...:', (targets['labels'])[i].item())
#print('Pred..: ' , (box_labels[box_closed[1]]).item())
if (targets['labels'])[i].item() == (box_labels[box_closed[1]]).item():
labels.append((box_labels[box_closed[1]]).item())
scores.append((box_scores[box_closed[1]]).item())
boxes.append(box_closed[0])
box_scores = [v for i,v in enumerate(box_scores) if box_closed[1] !=i]
box_labels = [v for i,v in enumerate(box_labels) if box_closed[1] !=i]
box_preds = [v for i,v in enumerate(box_preds) if box_closed[1] !=i]
boxes = [bx.cpu().tolist() for bx in boxes]
boxes_fp = [[bx.cpu().tolist(),box_labels[i].item(),round(box_scores[i].item(),3)] for i,bx in enumerate(box_preds) if box_scores[i].item() >= self.threshold_class]
preds_closed ={
'boxes' : torch.tensor(boxes),
'scores': torch.tensor(scores),
'labels': torch.tensor(labels),
}
if qt_show > 0:
tg_boxes = [ [value.cpu().tolist(),0,0] for value in targets['boxes']]
img = self.boxes_draw.draw_rectangle(bboxes=tg_boxes, img=img,dim=(640,640),color=(255,0,0))
cv2.imwrite('img_'+str(qt_show)+'.png',img)
bxs =[[bx,labels[i],round(scores[i],2)] for i,bx in enumerate(boxes)]
img = self.boxes_draw.draw_rectangle(bboxes=boxes_fp, img=img,dim=(640,640),color=(0,0,255), classes=classes)
img = self.boxes_draw.draw_rectangle(bboxes=b_gt, img=img,dim=(640,640),color=(255,0,0))
img = self.boxes_draw.draw_rectangle(bboxes=bxs, img=img,dim=(640,640), classes=classes)
img = self.boxes_draw.draw_text(img, 'A..:'+str(len(b_gt)), (10,15), color=(255,0,0))
img = self.boxes_draw.draw_text(img, 'Tp..:'+str(len(bxs)), (10,35), color=(0,255,0))
img = self.boxes_draw.draw_text(img, 'Fp..:'+str(len(boxes_fp)), (10,55), color=(0,0,255))
qt_show -= 1
cv2.imwrite(os.path.join(self.path,'img_'+str(qt_show)+'.png'),img)
qt_show -=1
self.metrics.update([preds_closed],[targets])
fp_t += len(boxes_fp)
tp_t += len(bxs)
total += len(b_gt)
metr = self.metrics.compute()
data = {}
data['Technique'] = self.technique
data['Fold'] = self.fold
data['Map'] = round(metr['map'].item(),3)
data['Map_50'] = round(metr['map_50'].item(),3)
data['Map_75'] = round(metr['map_75'].item(),3)
data['Tp'] = tp_t
data['Fp'] = fp_t
data['Precision'] = self.operation(numerator=tp_t,denominator=(tp_t+fp_t))
data['Recall'] = self.operation(numerator=tp_t, denominator=(tp_t+(total-(tp_t))))
data['Total'] = total
self.boxes_draw.save_file(data,os.path.join(self.estatistics_local,'statistics.csv'))
print("Testing executed !")
except ValueError as error:
print(error)
def operation(self, numerator, denominator, precision=3):
try:
value = round(numerator/float(denominator), precision)
return value
except ZeroDivisionError:
return 0.000
if __name__ == '__main__':
#model = init_detector('../work_dirs/ssd300_coco/ssd300_coco.py', 'work_dirs/ssd300_coco/best_coco_bbox_mAP_epoch_10.pth', device=torch.device('cuda:0'))
model = init_detector('checkpoints/ssd300_coco/ssd300_coco.py', 'checkpoints/ssd300_coco/best_coco_bbox_mAP_epoch_5.pth', device=torch.device('cuda:0'))
testing = Testing(
model=None,
prefix=os.path.join(os.getcwd(),'../../../dataset/all/train'),
path='../../../dataset/filesJSON/fold_1_test.json'
model = model,
results_path = '.',
prefix = os.path.join(os.getcwd(),'../../SUINDETEC/all/train'),
path_json ='../../SUINDETEC/filesJSON/fold_1_test.json'
)
testing.running()
\ No newline at end of file
testing.running()
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment