Skip to content

Commit

Permalink
LODet_v1.0
Browse files Browse the repository at this point in the history 
LODet_v1.0
  • Loading branch information
@Shank2358
Shank2358 committed Apr 29, 2021
1 parent 81d4a8f commit d6cec5f
Show file tree
Hide file tree
Showing 119 changed files with 15,913 additions and 0 deletions.
95 changes: 95 additions & 0 deletions config/cfg_lodet.py
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,95 @@
# coding=utf-8

PROJECT_PATH = "./"
DATA_PATH = "/mnt/Datasets/DIOR/"


DATA = {"CLASSES":['airplane','airport','baseballfield','basketballcourt','bridge','chimney',
'dam','Expressway-Service-area','Expressway-toll-station','golffield','groundtrackfield','harbor',
'overpass','ship','stadium','storagetank','tenniscourt','trainstation','vehicle','windmill'],
"NUM":20}
'''
MODEL = {
"ANCHORS":[[(1.494992296, 1.772419808), (2.550184278, 5.105188103), (4.511253175, 2.041398611)], # Anchors for small obj
[(3.852394468, 3.413543783), (3.827394513, 9.012606993), (7.569651633, 7.192874667)], # Anchors for medium obj
[(5.269568089, 8.068825014), (10.13079538, 3.44005408), (10.41848982, 10.60006263)]], # Anchors for big obj
"STRIDES":[8, 16, 32],
"ANCHORS_PER_SCLAE":3
}#544
'''
MODEL = {
"ANCHORS":[[(3.18524223, 1.57625129), (1.95394566,4.29178376), (6.65929852, 2.8841753)], # Anchors for small obj
[(1.9038, 4.42035), (6.712, 3.29255), (6.645, 12.7675)], # Anchors for medium obj
[(5.513875, 14.38123), (11.66746, 4.2333), (15.70345, 11.94367)]], # Anchors for big obj
"STRIDES":[8, 16, 32],
"ANCHORS_PER_SCLAE":3
}#800

MAX_LABEL = 500
SHOW_HEATMAP = False
SCALE_FACTOR=2.0

TRAIN = {
"EVAL_TYPE":'VOC', #['VOC', 'COCO']
"TRAIN_IMG_SIZE":800,
"TRAIN_IMG_NUM":11759,#11759,
"AUGMENT":True,
"MULTI_SCALE_TRAIN":True,
"MULTI_TRAIN_RANGE":[12,25,1],
"BATCH_SIZE":10,
"IOU_THRESHOLD_LOSS":0.5,
"EPOCHS":121,
"NUMBER_WORKERS":16,
"MOMENTUM":0.9,
"WEIGHT_DECAY":0.0005,
"LR_INIT":1.5e-4,
"LR_END":1e-6,
"WARMUP_EPOCHS":5,
"IOU_TYPE":'CIOU' #['GIOU','CIOU']
}

TEST = {
"EVAL_TYPE":'VOC', #['VOC', 'COCO', 'BOTH']
"EVAL_JSON":'test.json',
"EVAL_NAME":'test',
"NUM_VIS_IMG":0,
"TEST_IMG_SIZE":800,
"BATCH_SIZE":1,
"NUMBER_WORKERS":4,
"CONF_THRESH":0.05,
"NMS_THRESH":0.45,
"NMS_METHODS":'NMS', #['NMS', 'SOFT_NMS', 'NMS_DIOU', #'NMS_DIOU_SCALE']
"MULTI_SCALE_TEST":False,
"MULTI_TEST_RANGE":[320,640,96],
"FLIP_TEST":False
}


'''
DOTA_cfg
DATA = {"CLASSES": ['plane',
'baseball-diamond',
'bridge',
'ground-track-field',
'small-vehicle',
'large-vehicle',
'ship',
'tennis-court',
'basketball-court',
'storage-tank', 'soccer-ball-field', 'roundabout', 'harbor', 'swimming-pool', 'helicopter'],
"NUM": 15}
MODEL = {"ANCHORS":[[(1.625, 2.656), ( 3.652, 3.981), (4.493, 1.797)],
[(4.358,3.123), (2.000, 4.558), (6.077, 6.688)],
[(2.443, 7.848), (6.237, 4.750), (9.784, 10.291)]] ,# Anchors for big obj 608
"STRIDES":[8, 16, 32],
"ANCHORS_PER_SCLAE":3
}#544
MODEL = {"ANCHORS":[[(2.80340246, 2.87380792), (4.23121697, 6.44043634), (7.38428433, 3.82613533)],
[(4.2460819, 4.349495965), (4.42917327, 10.59395029), (8.24772929, 6.224761455)],
[(6.02687863, 5.92446062), (7.178407523, 10.86361071), (15.30253702, 12.62863728)]] ,# Anchors for big obj 608
"STRIDES":[8, 16, 32],
"ANCHORS_PER_SCLAE":3
}#800
'''
167 changes: 167 additions & 0 deletions dataR/evalR/coco_eval.py
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,167 @@
import json
import tempfile
import matplotlib.pyplot as plt
from torch.autograd import Variable
from torch.utils.data import DataLoader
from pycocotools.cocoeval import COCOeval
import time
from tqdm import tqdm
from dataload.cocodataset import *
from eval.evaluator import Evaluator
from utils.utils_coco import *
from utils.visualize import *

current_milli_time = lambda: int(round(time.time() * 1000))

class COCOEvaluator():
"""
COCO AP Evaluation class.
All the data in the val2017 dataset are processed \
and evaluated by COCO API.
"""
def __init__(self, data_dir, img_size, confthre, nmsthre):
"""
Args:
model_type (str): model name specified in config file
data_dir (str): dataset root directory
img_size (int): image size after preprocess. images are resized \
to squares whose shape is (img_size, img_size).
confthre (float):
confidence threshold ranging from 0 to 1, \
which is defined in the config file.
nmsthre (float):
IoU threshold of non-max supression ranging from 0 to 1.
"""
self.classes = cfg.DATA["CLASSES"]
self.val_data_path = cfg.DATA_PATH
self.pred_result_path = os.path.join(cfg.PROJECT_PATH, 'prediction')
self.__visual_imgs = cfg.TEST["NUM_VIS_IMG"]

augmentation = {'LRFLIP': False, 'JITTER': 0, 'RANDOM_PLACING': False,
'HUE': 0, 'SATURATION': 0, 'EXPOSURE': 0, 'RANDOM_DISTORT': False}

self.dataset = COCODataset(data_dir=data_dir,
img_size=img_size,
augmentation=augmentation,
json_file=cfg.TEST["EVAL_JSON"],
name=cfg.TEST["EVAL_NAME"])
self.dataloader = DataLoader(self.dataset, batch_size=cfg.TEST["BATCH_SIZE"], shuffle=False,
pin_memory=True, num_workers=cfg.TEST["NUMBER_WORKERS"])
self.img_size = img_size
self.confthre = confthre
self.nmsthre = nmsthre
self.inference_time = 0.
def evaluate(self, model):
"""
COCO average precision (AP) Evaluation. Iterate inference on the test dataset
and the results are evaluated by COCO API.
Args:
model : model object
Returns:
ap50_95 (float) : calculated COCO AP for IoU=50:95
ap50 (float) : calculated COCO AP for IoU=50
"""
model.eval()
cuda = torch.cuda.is_available()
Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
ids = []
data_dict = []
dataiterator = iter(self.dataloader)
#print(" Val datasets number is : {}".format(len(self.dataloader)))
for i in tqdm(range(len(self.dataloader))):
#while True:
#try:
img, _, info_img, id_, img_path = next(dataiterator) # load a batch
#except StopIteration:
#break
info_img = [float(info.numpy()) for info in info_img]
id_ = int(id_)
ids.append(id_)
with torch.no_grad():
img = Variable(img.type(Tensor))
start_time = current_milli_time()
_,outputs = model(img)
self.inference_time += (current_milli_time() - start_time)
outputs=outputs.unsqueeze(0)
outputs = postprocess(
outputs, cfg.DATA["NUM"], self.confthre, self.nmsthre)
if outputs[0] is None:
continue
outputs = outputs[0].cpu().data

for output in outputs:
x1 = float(output[0])
y1 = float(output[1])
x2 = float(output[2])
y2 = float(output[3])
label = self.dataset.class_ids[int(output[6])]
box = box2label((y1, x1, y2, x2), info_img)
bbox = [box[1], box[0], box[3] - box[1], box[2] - box[0]]
score = float(output[4].data.item() * output[5].data.item()) # object score * class score
A = {"image_id": id_, "category_id": label, "bbox": bbox,
"score": score, "segmentation": []} # COCO json format
data_dict.append(A)

if self.__visual_imgs and i <= self.__visual_imgs:
imgshow = cv2.imread(img_path[0])
bboxes_prd = Evaluator(model).get_bbox(imgshow, cfg.TEST["MULTI_SCALE_TEST"], cfg.TEST["FLIP_TEST"])
if bboxes_prd.shape[0] != 0:
boxes = bboxes_prd[..., :4]
class_inds = bboxes_prd[..., 5].astype(np.int32)
scores = bboxes_prd[..., 4]
visualize_boxes(image=imgshow, boxes=boxes, labels=class_inds, probs=scores, class_labels=self.classes)
path = os.path.join(self.pred_result_path, "imgs/{}.jpg".format(i))
cv2.imwrite(path, imgshow)


annType = ['segm', 'bbox', 'keypoints']
self.inference_time = 1.0 * self.inference_time / len(self.dataloader)
# Evaluate the Dt (detection) json comparing with the ground truth
if len(data_dict) > 0:
cocoGt = self.dataset.coco
_, tmp = tempfile.mkstemp()
json.dump(data_dict, open(tmp, 'w'))
cocoDt = cocoGt.loadRes(tmp)
cocoEval = COCOeval(self.dataset.coco, cocoDt, annType[1])
cocoEval.params.imgIds = ids
cocoEval.evaluate()
cocoEval.accumulate()
cocoEval.summarize()

'''
# ----------pltshow------------- #
# precision[t,:,k,a,m] PR curves recall-precision value
# T:IoU thresh.5-.95, gap=0.05, t[0]=0.5,t[1]=0.55,t[2]=0.6,t[3]=0.65,t[4]=0.7,t[5]=0.75 ……,t[9]=0.95
# R:101 recall thresh，0-101
# K:class k[0] = person,k[1] = bycicle,.....COCO
# A:area, a[0]=all,a[1]=small,a[2]=medium,a[3]=large
# M:Maxdet m[0]=1,m[1]=10,m[2]=100
#C75: PR at IoU=.75 (AP at strict IoU), area under curve corresponds to APIoU=.75 metric.
#C50: PR at IoU=.50 (AP at PASCAL IoU), area under curve corresponds to APIoU=.50 metric.
#Loc: PR at IoU=.10 (localization errors ignored, but not duplicate detections). All remaining settings use IoU=.1.
#Sim: PR after supercategory false positives (fps) are removed. Specifically, any matches to objects with a different class label but that belong to the same supercategory don't count as either a fp (or tp). Sim is computed by setting all objects in the same supercategory to have the same class label as the class in question and setting their ignore flag to 1. Note that person is a singleton supercategory so its Sim result is identical to Loc.
#Oth: PR after all class confusions are removed. Similar to Sim, except now if a detection matches any other object it is no longer a fp (or tp). Oth is computed by setting all other objects to have the same class label as the class in question and setting their ignore flag to 1.
#BG: PR after all background (and class confusion) fps are removed. For a single category, BG is a step function that is 1 until max recall is reached then drops to 0 (the curve is smoother after averaging across categories).
#FN: PR after all remaining errors are removed (trivially AP=1).
pr_array1 = cocoEval.eval['precision'][0, :, 0, 0, 2]
pr_array2 = cocoEval.eval['precision'][5, :, 0, 0, 2]
#pr_array3 = cocoEval.eval['precision'][6, :, 0, 0, 2]
#pr_array4 = cocoEval.eval['precision'][9, :, 0, 0, 2]
x = np.arange(0.0, 1.01, 0.01)
# x_1 = np.arange(0, 1.01, 0.111)
plt.xlabel('IoU')
plt.ylabel('precision')
plt.xlim(0, 1.0)
plt.ylim(0, 1.01)
plt.grid(True)
plt.plot(x, pr_array1, color='blue', linewidth = '3', label='IoU=0.5')
plt.plot(x, pr_array2, color='green', linewidth = '3', label='IoU=0.75')
plt.title("P-R curves catid=person maxDet=100")
plt.legend(loc="lower left")
plt.savefig("../prediction/APs.png", dpi=600)
# plt.show()'''
return cocoEval.stats[0], cocoEval.stats[1], self.inference_time
else:
return 0, 0, 0
Loading

0 comments on commit d6cec5f

Please sign in to comment.