def generate_prob_map_hr(cfg, file_name, sorted_abnormal_patches, model_hr,
time, patch_out_size, b_map, p_map):
# for high resolution image.
t = time
slide = extract_patch_fun.single_img_process(file_name, None, None, None,
False)
gm_dataset = dataloader_fun.gm_cls_DataLoader_hr(sorted_abnormal_patches,
slide._img, 0)
gm_loader = torch.utils.data.DataLoader(gm_dataset,
batch_size=cfg.gm_batch_size,
shuffle=False,
num_workers=cfg.gm_work_num)
print('start inference the model')
output = _get_label_prob(gm_loader, model_hr)
print('Done! %.4fs' % t.value())
output_b = np.argmax(output, axis=1)
output_p = output[:, 1]
print('start fill the output into the map')
b_map = _fill_list_into_map(sorted_abnormal_patches, b_map, output_b,
patch_out_size)
p_map = _fill_list_into_map(sorted_abnormal_patches, p_map, output_p,
patch_out_size)
print('Done! %.4fs' % t.value())
return b_map, p_map
def generate_prob_map(cfg, model, file_name):
t = timemeter.TimeMeter()
slide = extract_patch_fun.single_img_process(file_name, None, None, None,
False)
print('start extract background ')
img, mask = slide._generate_img_bg_mask()
print('Done! %.4fs' % t.value())
size_raw = slide._img.level_dimensions[0]
mask_frac = size_raw[1] * 1.0 / mask.shape[0]
# calculate the gm row column and patch out size
row = cal_new_len(size_raw[1])
col = cal_new_len(size_raw[0])
patch_out_size = cal_new_len(cfg.windows_size)
size_out = (col, row)
img = img.resize(size_out)
# mask = _np_resize(mask, (size_out[1], size_out[0]))
b_map = np.zeros((row, col)).astype(np.uint8)
p_map = np.zeros((row, col)).astype(np.float32)
print('start get input list ')
input_list = _get_input_list(mask, mask_frac, cfg.windows_size, size_raw,
size_out, patch_out_size)
print('Done! %.4fs' % t.value())
print('get %d input patch' % len(input_list))
gm_dataset = dataloader_fun.gm_fcn_DataLoader(input_list, slide._img,
cfg.windows_size)
gm_loader = torch.utils.data.DataLoader(gm_dataset,
batch_size=cfg.gm_batch_size,
shuffle=False,
num_workers=cfg.gm_work_num)
print('start inference the model')
output = _get_label_prob(gm_loader, model)
print('Done! %.4fs' % t.value())
output_b = np.argmax(output, axis=1)
output_p = output[:, 1]
print('start fill the output into the map')
b_map = _fill_list_into_map(input_list, b_map, output_b, patch_out_size)
p_map = _fill_list_into_map(input_list, p_map, output_p, patch_out_size)
print('Done! %.4fs' % t.value())
return img, b_map, p_map
def patch_filter(cfg, model, file_name, level):
# for low resolution.
t = timemeter.TimeMeter()
slide = extract_patch_fun.single_img_process(file_name, None, None, None,
False)
print('start extract background ')
img, mask = slide._generate_img_bg_mask_ForTest()
print('Done! %.4fs' % t.value())
mask_frac = slide._times_target_level_divide_rescaled_mask
size_raw = slide._img.level_dimensions[level]
scale_time = slide._rescaled_times
patch_out_size = cal_new_len_lr(cfg.patch_size, mask_frac)
row = cal_new_len_lr(size_raw[1], mask_frac)
col = cal_new_len_lr(size_raw[0], mask_frac)
b_map = np.zeros((row, col)).astype(np.uint8)
p_map = np.zeros((row, col)).astype(np.float32)
#row = cal_new_len_lr(size_raw[1], mask_frac)
#col = cal_new_len_lr(size_raw[0], mask_frac)
print('start get input list ')
input_list = _get_input_list_lr(mask, cfg.patch_size, size_raw, scale_time,
mask_frac, patch_out_size)
print('Done! %.4fs' % t.value())
print('get %d input patch' % len(input_list))
gm_dataset = dataloader_fun.gm_cls_DataLoader(input_list, slide._img,
cfg.patch_size, level)
gm_loader = torch.utils.data.DataLoader(gm_dataset,
batch_size=cfg.gm_batch_size,
shuffle=False,
num_workers=cfg.gm_work_num)
print('start inference the model')
sorted_abnormal_patches = _sort_abnormal_patches(gm_loader, model,
input_list)
print('Done! %.4fs' % t.value())
return img, sorted_abnormal_patches, t, patch_out_size, b_map, p_map
def generate_prob_map(cfg, model, file_name):
t = timemeter.TimeMeter()
slide = extract_patch_fun.single_img_process(file_name, None, None, None, False)
print('start extract background ')
img, mask = slide._generate_img_bg_mask()
print('Done! %.4fs'%t.value())
size_raw = slide._img.level_dimensions[0]
size_out = (int(size_raw[0]/cfg.gm_stride), int(size_raw[1]/cfg.gm_stride))
frac = np.ceil(size_raw[0]/size_out[0])
img = img.resize(size_out)
mask = _np_resize(mask, size_out)
b_map = np.zeros(mask.shape).astype(np.uint8)
p_map = np.zeros(mask.shape).astype(np.float32)
print('start get input list ')
input_list = _get_input_list(cfg, mask, frac)
print('Done! %.4fs' % t.value())
print('get %d input patch'%len(input_list))
gm_dataset = dataloader_fun.gm_cls_DataLoader(input_list, slide._img, cfg.patch_size)
gm_loader = torch.utils.data.DataLoader(gm_dataset, batch_size=cfg.gm_batch_size,
shuffle=False, num_workers=cfg.gm_work_num)
print('start inference the model')
output = _get_label_prob(gm_loader, model)
print('Done! %.4fs' % t.value())
output_b = np.argmax(output, axis=1)
output_p = output[:, 1]
print('start fill the output into the map')
b_map = _fill_list_into_map(input_list, b_map, output_b)
p_map = _fill_list_into_map(input_list, p_map, output_p)
print('Done! %.4fs' % t.value())
return img, b_map, p_map