def validate(self):
model.eval()
with torch.no_grad():
iou_epoch = 0.
iou_NN_epoch = 0.
num_batches = 0
loss_epoch = 0.
# Validation loop
for i, data in enumerate(tqdm(dataloader_val), 0):
# data creation
tgt_odms = data['odms'].to(args.device)
tgt_voxels = data['voxels'].to(args.device)
inp_voxels = down_sample(tgt_voxels)
inp_odms = []
for voxel in inp_voxels:
inp_odms.append(kal.rep.voxel.extract_odms(voxel).unsqueeze(0))
inp_odms = torch.cat(inp_odms)
tgt_odms_occ = to_occpumancy_map(tgt_odms)
# inference
pred_odms = model(inp_odms)
# losses
loss = loss_fn(pred_odms, tgt_odms_occ)
loss_epoch += float(loss.item())
ones = pred_odms > .5
zeros = pred_odms <= .5
pred_odms[ones] = pred_odms.shape[-1]
pred_odms[zeros] = 0
NN_pred = up_sample(inp_voxels)
iou_NN = kal.metrics.voxel.iou(NN_pred.contiguous(), tgt_voxels)
iou_NN_epoch += iou_NN
pred_voxels = []
for odms, voxel_NN in zip(pred_odms, NN_pred):
pred_voxels.append(kal.rep.voxel.project_odms(odms, voxel = voxel_NN, votes = 2).unsqueeze(0))
pred_voxels = torch.cat(pred_voxels)
iou = kal.metrics.voxel.iou(pred_voxels.contiguous(), tgt_voxels)
iou_epoch += iou
# logging
num_batches += 1
if i % args.print_every == 0:
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(f'[VAL] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}')
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(f'[VAL Total] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}')
loss_epoch = loss_epoch / num_batches
self.val_loss.append(out_iou)
def validate(self):
model.eval()
with torch.no_grad():
iou_epoch = 0.
iou_NN_epoch = 0.
num_batches = 0
loss_epoch = 0.
# Validation loop
for i, sample in enumerate(tqdm(dataloader_val), 0):
data = sample['data']
# data creation
tgt = data['128'].to(device)
inp = data['32'].to(device)
# inference
pred = model(inp.unsqueeze(1))
# losses
tgt = tgt.long()
loss = loss_fn(pred, tgt)
loss_epoch += float(loss.item())
pred = pred[:, 1, :, :]
iou = kal.metrics.voxel.iou(pred.contiguous(), tgt)
iou_epoch += iou
NN_pred = up_sample(inp)
iou_NN = kal.metrics.voxel.iou(NN_pred.contiguous(), tgt)
iou_NN_epoch += iou_NN
# logging
num_batches += 1
if i % args.print_every == 0:
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(
f'[VAL] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}'
)
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(
f'[VAL Total] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}'
)
loss_epoch = loss_epoch / num_batches
self.val_loss.append(out_iou)
iou_NN_epoch = 0.
num_batches = 0
model.eval()
with torch.no_grad():
for data in tqdm(dataloader_val):
tgt = data['data'].to(args.device)
inp = down_sample(tgt)
# inference
pred = model(inp)
iou = kal.metrics.voxel.iou(pred[:,1,:,:].contiguous(), tgt)
iou_epoch += iou
NN_pred = up_sample(inp)
iou_NN = kal.metrics.voxel.iou(NN_pred.contiguous(), tgt)
iou_NN_epoch += iou_NN
if args.vis:
for i in range(pred.shape[0]):
print ('Rendering low resolution input')
kal.visualize.show_voxel(inp[i,0], mode = 'exact', thresh = .5)
print ('Rendering high resolution target')
kal.visualize.show_voxel(tgt[i], mode = 'exact', thresh = .5)
print ('Rendering high resolution prediction')
kal.visualize.show_voxel(pred[i,1], mode = 'exact', thresh = .5)
print('----------------------')
num_batches += 1.
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
iou_NN_epoch = 0.
num_batches = 0
model.eval()
with torch.no_grad():
for data in tqdm(dataloader_val):
tgt_odms = data['odms_128'].to(args.device)
tgt_voxels = data['voxels_128'].to(args.device)
inp_odms = data['odms_32'].to(args.device)
inp_voxels = data['voxels_32'].to(args.device)
# inference
pred_odms = model(inp_odms) * 128
NN_pred = up_sample(inp_voxels)
iou_NN = kal.metrics.voxel.iou(NN_pred.contiguous(), tgt_voxels)
iou_NN_epoch += iou_NN
pred_odms = pred_odms.int()
pred_voxels = []
for odms, NN_odms in zip(pred_odms, NN_pred):
pred_voxels.append(
kal.rep.voxel.project_odms(odms, voxel=NN_odms,
votes=2).unsqueeze(0))
pred_voxels = torch.cat(pred_voxels)
iou = kal.metrics.voxel.iou(pred_voxels.contiguous(), tgt_voxels)
iou_epoch += iou
if args.vis:
def validate(self):
model.eval()
with torch.no_grad():
iou_epoch = 0.
iou_NN_epoch = 0.
num_batches = 0
loss_epoch = 0.
# Validation loop
for i, data in enumerate(tqdm(dataloader_val), 0):
# data creation
tgt_odms = data['odms_128'].to(args.device)
tgt_voxels = data['voxels_128'].to(args.device)
inp_odms = data['odms_32'].to(args.device)
inp_voxels = data['voxels_32'].to(args.device)
# inference
initial_odms = upsample_omd(inp_odms) * 4
distance = 128 - initial_odms
pred_odms_update = model(inp_odms)
pred_odms_update = pred_odms_update * distance
pred_odms = initial_odms + pred_odms_update
# losses
loss = loss_fn(pred_odms, tgt_odms)
loss_epoch += float(loss.item())
NN_pred = up_sample(inp_voxels)
iou_NN = kal.metrics.voxel.iou(NN_pred.contiguous(),
tgt_voxels)
iou_NN_epoch += iou_NN
pred_voxels = []
pred_odms = pred_odms.int()
for odms, voxel_NN in zip(pred_odms, NN_pred):
pred_voxels.append(
kal.rep.voxel.project_odms(odms, voxel_NN,
votes=2).unsqueeze(0))
pred_voxels = torch.cat(pred_voxels)
iou = kal.metrics.voxel.iou(pred_voxels.contiguous(),
tgt_voxels)
iou_epoch += iou
# logging
num_batches += 1
if i % args.print_every == 0:
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(
f'[VAL] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}'
)
out_iou = iou_epoch.item() / float(num_batches)
out_iou_NN = iou_NN_epoch.item() / float(num_batches)
tqdm.write(
f'[VAL Total] Epoch {self.cur_epoch:03d}, Batch {i:03d}: IoU: {out_iou}, Iou Base: {out_iou_NN}'
)
loss_epoch = loss_epoch / num_batches
self.val_loss.append(out_iou)
from model import FFTSR
import tensorflow as tf
import numpy as np
import cv2
from utils import fft, bicubic, up_sample, imshow, ifft, imshow_spectrum
from matplotlib import pyplot as plt
if __name__ == '__main__':
img = 'images_train/butterfly.bmp'
# img = cv2.imread(img,cv2.IMREAD_GRAYSCALE)
img = cv2.imread(img)
img = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)
print('img_shape ->', img.shape) #Y cr cb
# img = img.reshape([1,256,256,1])
with tf.Session() as sess:
hr_img = (img) / 255.0 * (1e3 * 1e-5)
lr_img = (up_sample(bicubic(img))) / 255.0 * (1e3 * 1e-5)
# imshow_spectrum(lr_img)
fftsr = FFTSR(sess, 1e-4, 10000)
# fftsr.build_model()
# fftsr.run(hr_img,lr_img)
fftsr.run(hr_img[:, :, 0], lr_img[:, :, 0])
# out = fftsr.pred
# print(out)