def valid_calc(img_model, txt_model, dataset, bit, batch_size, drop_integer=False, return_hash=False, cuda=True):
"""
get valid data hash code and calculate mAP
:param img_model: the image model
:param txt_model: the txt model
:param dataset: the valid dataset
:param bit: the length of hash code
:param batch_size: the batch size of valid
:param drop_integer: if true, the excrescent data will be drop
:param return_hash: if true, the hash codes will be returned
:param cuda: if use cuda
:return: mAP and hash codes(if return_hash = True)
"""
# get query img and txt binary code
dataset.query()
qB_img, qB_txt = TrainBase.get_codes(img_model, txt_model, dataset, bit, batch_size, cuda=cuda)
query_label = dataset.get_all_label()
# get retrieval img and txt binary code
dataset.retrieval()
rB_img, rB_txt = TrainBase.get_codes(img_model, txt_model, dataset, bit, batch_size, cuda=cuda)
retrieval_label = dataset.get_all_label()
mAPi2t = calc_map_k(qB_img, rB_txt, query_label, retrieval_label)
mAPt2i = calc_map_k(qB_txt, rB_img, query_label, retrieval_label)
mAPi2i = calc_map_k(qB_img, rB_img, query_label, retrieval_label)
mApt2t = calc_map_k(qB_txt, rB_txt, query_label, retrieval_label)
if return_hash:
return mAPi2t, mAPt2i, mAPi2i, mApt2t, qB_img.cpu(), qB_txt.cpu(), rB_img.cpu(), rB_txt.cpu()
return mAPi2t, mAPt2i, mAPi2i, mApt2t
def calc_map(ind):
qB_img_ind = qB_img[:, ind]
qB_txt_ind = qB_txt[:, ind]
rB_img_ind = rB_img[:, ind]
rB_txt_ind = rB_txt[:, ind]
mAPi2t = calc_map_k(qB_img_ind, rB_txt_ind, query_label,
retrieval_label)
mAPt2i = calc_map_k(qB_txt_ind, rB_img_ind, query_label,
retrieval_label)
return mAPi2t, mAPt2i
def calc_map(qB_img, qB_txt, rB_img, rB_txt, query_label, retrieval_label):
mAPi2t = calc_map_k(qB_img, rB_txt, query_label, retrieval_label)
mAPt2i = calc_map_k(qB_txt, rB_img, query_label, retrieval_label)
return mAPi2t, mAPt2i
