def test_coco(model):
global oh, ow
catId2cls, cls2catId, catId2name = joblib.load(
r'D:/(catId2cls,cls2catId,catId2name).pkl')
test_dir = r'D:\dataset\val2017/'
names = os.listdir(test_dir)
names = [name.split('.')[0] for name in names]
names = sorted(names)
i = 0
mm = 1000000
Res = []
Res_mask = []
start_time = datetime.now()
for name in names[:mm]:
i += 1
print(datetime.now(), i)
im_file = test_dir + name + '.jpg'
img = cv2.imread(im_file)
oh, ow = img.shape[:2]
with torch.no_grad():
res = model.predict(img)
res = res[:200]
wh = res[:, 2:4] - res[:, :2] + 1
imgId = int(name)
m = res.shape[0]
imgIds = np.zeros((m, 1)) + imgId
cls = res[:, 5]
cid = map(lambda x: cls2catId[x], cls)
cid = list(cid)
cid = np.array(cid)
cid = cid.reshape(-1, 1)
res = np.concatenate((imgIds, res[:, :2], wh, res[:, 4:5], cid),
axis=1)
# Res=np.concatenate([Res,res])
res = np.round(res, 4)
Res.append(res)
Res = np.concatenate(Res, axis=0)
Ann = loadNumpyAnnotations(Res)
print('==================================', mm,
datetime.now() - start_time)
with codecs.open('yolov3_bbox.json', 'w', 'ascii') as f:
json.dump(Ann, f)
eval_coco_box.eval('yolov3_bbox.json', mm)
pass
def test(model, config, model_file):
global oh, ow
catId2cls, cls2catId, catId2name = joblib.load(
r'../data_preprocess/(catId2cls,cls2catId,catId2name).pkl')
model = model(config)
model.eval()
model.load_state_dict(torch.load(model_file, map_location='cpu'))
cuda(model)
test_dir = r'D:/dataset/val2017/'
names = os.listdir(test_dir)
names = [name.split('.')[0] for name in names]
names = sorted(names)
i = 0
mm = 10000000
Res = []
Res_mask = []
start_time = datetime.now()
for name in names[:mm]:
i += 1
print(datetime.now(), i)
im_file = test_dir + name + '.jpg'
img = cv2.imread(im_file)
oh, ow = img.shape[:2]
with torch.no_grad():
img = torch.tensor(img)
res, res_mask = model(img)
res = res.cpu().numpy()
# res = res.detach().numpy()
res_mask = res_mask.cpu()
res_mask = res_mask.detach().numpy()
wh = res[:, 2:4] - res[:, :2]
imgId = int(name)
m = res.shape[0]
imgIds = np.zeros((m, 1)) + imgId
cls = res[:, 5]
cid = map(lambda x: cls2catId[x], cls)
cid = list(cid)
cid = np.array(cid)
cid = cid.reshape(-1, 1)
res = np.concatenate((imgIds, res[:, :2] + 1, wh, res[:, 4:5], cid),
axis=1)
# Res=np.concatenate([Res,res])
res = np.round(res, 4)
Res.append(res)
Res_mask += map(loadNumpyAnnotations_mask, res[:100], res_mask[:100])
Res = np.concatenate(Res, axis=0)
Ann = loadNumpyAnnotations(Res)
print('==================================', mm,
datetime.now() - start_time)
# with codecs.open('Mask_Rcnn_bbox_ohem_256_gpu2.json', 'w', 'ascii') as f:
# json.dump(Ann, f)
# with codecs.open('Mask_Rcnn_segm_ohem_256_gpu2.json', 'w', 'ascii') as f:
# json.dump(Res_mask, f)
# eval_coco_box.eval('Mask_Rcnn_bbox_ohem_256_gpu2.json', mm)
# eval_coco_segm.eval('Mask_Rcnn_segm_ohem_256_gpu2.json', mm)
with codecs.open('py_Mask_Rcnn_bbox.json', 'w', 'ascii') as f:
json.dump(Ann, f)
with codecs.open('py_Mask_Rcnn_segm.json', 'w', 'ascii') as f:
json.dump(Res_mask, f)
print(mm)
eval_coco_box.eval('py_Mask_Rcnn_bbox.json', mm)
eval_coco_segm.eval('py_Mask_Rcnn_segm.json', mm)
print('==========', datetime.now() - start_time)