labelIndexBatch = Variable(torch.LongTensor(opt.batchSize, 1, 300, 300))
# Initialize network
encoder_normal = model.encoder()
decoder_normal = model.decoder()
model_root_normal = '/datasets/cse152-252-sp20-public/unet_checkpoints/unet_original_zq'
epoch_id_normal = 181
encoder_normal.load_state_dict(
torch.load('%s/encoder_%d.pth' % (model_root_normal, epoch_id_normal)))
decoder_normal.load_state_dict(
torch.load('%s/decoder_%d.pth' % (model_root_normal, epoch_id_normal)))
encoder_normal = encoder_normal.eval()
decoder_normal = decoder_normal.eval()
encoder_dilation = model.encoderDilation()
decoder_dilation = model.decoderDilation()
model_root_dilation = '/datasets/cse152-252-sp20-public/unet_checkpoints/unet_original_zq_dilation'
epoch_id_dilation = 180
encoder_dilation.load_state_dict(
torch.load('%s/encoder_%d.pth' % (model_root_dilation, epoch_id_dilation)))
decoder_dilation.load_state_dict(
torch.load('%s/decoder_%d.pth' % (model_root_dilation, epoch_id_dilation)))
encoder_dilation = encoder_dilation.eval()
decoder_dilation = decoder_dilation.eval()
encoder_spp = model.encoderDilation()
decoder_spp = model.decoderDilation(isSpp=True)
model_root_spp = '/datasets/cse152-252-sp20-public/unet_checkpoints/unet_original_zq_spp'
epoch_id_spp = 180
encoder_spp.load_state_dict(
torch.load('%s/encoder_%d.pth' % (model_root_spp, epoch_id_spp)))
os.system('mkdir %s' % opt.experiment )
os.system('cp *.py %s' % opt.experiment )
if torch.cuda.is_available() and opt.noCuda:
print("WARNING: You have a CUDA device, so you should probably run with --cuda")
# Initialize image batch
imBatch = Variable(torch.FloatTensor(opt.batchSize, 3, 300, 300) )
labelBatch = Variable(torch.FloatTensor(opt.batchSize, opt.numClasses, 300, 300) )
maskBatch = Variable(torch.FloatTensor(opt.batchSize, 1, 300, 300) )
labelIndexBatch = Variable(torch.LongTensor(opt.batchSize, 1, 300, 300) )
# Initialize network
if opt.isDilation:
encoder = model.encoderDilation()
decoder = model.decoderDilation()
elif opt.isSpp:
encoder = model.encoderSPP()
decoder = model.decoderSPP()
else:
encoder = model.encoder()
decoder = model.decoder()
encoder.load_state_dict(torch.load('%s/encoder_%d.pth' % (opt.modelRoot, opt.epochId) ) )
decoder.load_state_dict(torch.load('%s/decoder_%d.pth' % (opt.modelRoot, opt.epochId) ) )
encoder = encoder.eval()
decoder = decoder.eval()
# Move network and containers to gpu
if not opt.noCuda:
imBatch = imBatch.cuda(opt.gpuId )
# Initialize image batch
imBatch = Variable(
torch.FloatTensor(opt.batchSize, 3, opt.imHeight, opt.imWidth))
labelBatch = Variable(
torch.FloatTensor(opt.batchSize, opt.numClasses, opt.imHeight,
opt.imWidth))
maskBatch = Variable(
torch.FloatTensor(opt.batchSize, 1, opt.imHeight, opt.imWidth))
labelIndexBatch = Variable(
torch.LongTensor(opt.batchSize, 1, opt.imHeight, opt.imWidth))
# Initialize network
if opt.isDilation:
encoder = model.encoderDilation()
decoder = model.decoderDilation()
# decoder = model.decoder()
elif opt.isSpp:
encoder = model.encoderDilation()
decoder = model.decoderDilation(isSpp=True)
else:
encoder = model.encoder()
decoder = model.decoder()
if opt.isPretrained:
model.loadPretrainedWeight(encoder, isOutput=True)
# Move network and containers to gpu
if not opt.noCuda:
imBatch = imBatch.cuda(opt.gpuId)
labelBatch = labelBatch.cuda(opt.gpuId)
labelIndexBatch = labelIndexBatch.cuda(opt.gpuId)