def ReconstructSpatialLFPatch(LFpatch, model, inputs, is_training, session, args, stride=60, border=(3, 3)):
"""
Spatial reconstruct for a LF patch.
:param LFpatch: input LF patch
:param model: network
:param inputs: inputs of the network
:param is_training: scala tensor to indicate whether need training
:param session: tensorflow session
:param args: arguments
:param stride: stride for reconstruction
:param border: shaved border of the final reconstructed LF patch
:return: reconstructed LF patch (border shaved)
"""
recons_indices = get_indices(LFpatch, rs=args.gamma_S, patchsize=args.patchSize, stride=stride)
inPatches, _ = LF_split_patches(LFpatch, patchsize=args.patchSize, stride=stride)
reconPatches = []
for i in tqdm(range(len(inPatches))):
in_patch = inPatches[i]
recon_patch = session.run(model.Recons, feed_dict={inputs: in_patch, is_training: False})
reconPatches.append(recon_patch)
reconsPatch_shaved = shaved_LF_reconstruct(reconPatches, recons_indices, border=border)
reconsPatch_shaved[reconsPatch_shaved > 1.] = 1.
reconsPatch_shaved[reconsPatch_shaved < 0.] = 0.
return reconsPatch_shaved
def get_indices(data, rs=4, patchsize=96, stride=30):
"""
Calculate the indices for the splited LF patches.
:param data: input LF data
:param rs: upsampling factor
:param patchsize: split size
:param stride: stride for spliting
:return: indices for all splited LF patches
"""
b, h, w, s, t, c = data.shape
recons_template = np.zeros([b, h*rs, w*rs, s, t, c])
_, indices = LF_split_patches(recons_template, patchsize=patchsize, stride=stride)
return indices