labelflag="lod0",
weightflag="iou",
dataflag="train",
POTSDAM=False)
if sys.argv[1] == "POTSDAM":
data = segsemdata.makeISPRS(datasetpath=root + "ISPRS_POTSDAM",
dataflag="train",
POTSDAM=True)
if sys.argv[1] == "POTSDAM_lod0":
data = segsemdata.makeISPRS(datasetpath=root + "ISPRS_POTSDAM",
labelflag="lod0",
weightflag="iou",
dataflag="train",
POTSDAM=True)
if sys.argv[1] == "BRUGES":
data = segsemdata.makeDFC2015(datasetpath=root + "DFC2015",
dataflag="train")
if sys.argv[1] == "BRUGES_lod0":
data = segsemdata.makeDFC2015(datasetpath=root + "DFC2015",
labelflag="lod0",
weightflag="iou",
dataflag="train")
if sys.argv[1] == "TOULOUSE":
data = segsemdata.makeSEMCITY(datasetpath=root + "SEMCITY_TOULOUSE",
dataflag="train")
if sys.argv[1] == "TOULOUSE_lod0":
data = segsemdata.makeSEMCITY(datasetpath=root + "SEMCITY_TOULOUSE",
dataflag="train",
labelflag="lod0",
weightflag="iou")
if sys.argv[1] == "AIRS":
data = segsemdata.makeAIRSdataset(datasetpath=root + "AIRS",
device = "cuda" if torch.cuda.is_available() else "cpu"
if device == "cuda":
torch.cuda.empty_cache()
cudnn.benchmark = True
import segsemdata
import embedding
import numpy as np
print("load model")
net = embedding.Embedding(pretrained="/data/vgg16-00b39a1b.pth")
net = net.to(device)
print("load data")
datatrain = segsemdata.makeDFC2015(datasetpath="/data/DFC2015",
lod0=False,
dataflag="train")
datatrain = datatrain.copyTOcache(outputresolution=50)
net.adddataset(datatrain.metadata())
net = net.to(device)
nbclasses = len(datatrain.setofcolors)
earlystopping = datatrain.getrandomtiles(1000, 128, 16)
print("train setting")
import torch.nn as nn
import collections
import random
from sklearn.metrics import confusion_matrix
criterion = nn.CrossEntropyLoss()
optimizer = net.getoptimizer()
import segsemdata
import embedding
import numpy as np
print("load model")
net = embedding.Embedding(pretrained="/data/vgg16-00b39a1b.pth")
net = net.to(device)
print("load data")
datatrain1 = segsemdata.makeISPRS(datasetpath="/data/ISPRS_VAIHINGEN",
POTSDAM=False)
datatrain1 = datatrain1.copyTOcache(outputresolution=50)
net.adddataset(datatrain1.metadata())
datatrain2 = segsemdata.makeDFC2015()
datatrain2 = datatrain2.copyTOcache(outputresolution=50)
net.adddataset(datatrain2.metadata())
net = net.to(device)
earlystopping1 = datatrain1.getrandomtiles(1000, 128, 16)
earlystopping2 = datatrain2.getrandomtiles(1000, 128, 16)
print("train setting")
import torch.nn as nn
import collections
import random
from sklearn.metrics import confusion_matrix
criterion = nn.CrossEntropyLoss()
print("TRAIN AIRS+POST+DFC")
import torch
import segsemdata
import numpy as np
import torch.backends.cudnn as cudnn
import random
device = "cuda" if torch.cuda.is_available() else "cpu"
print("load data")
datatrain = segsemdata.MergedSegSemDataset(
[
segsemdata.makeTrainAIRSdataset(),
segsemdata.makeISPRS_POSTDAM(),
segsemdata.makeDFC2015(),
]
)
earlystopping = datatrain.getrandomtiles(25, 128, 128, 16)
print("load model")
net = segsemdata.Unet(2)
net.loadpretrained("/data/vgg16-00b39a1b.pth")
net = net.to(device)
if device == "cuda":
torch.cuda.empty_cache()
cudnn.benchmark = True
print("train setting")
import torch.nn as nn
import torch.optim as optim
import collections
_, pred = torch.max(pred[0], 0)
pred = pred.cpu().numpy()
assert (label.shape == pred.shape)
cm += confusion_matrix(label.flatten(), pred.flatten(),
list(range(nbclasses)))
pred = PIL.Image.fromarray(datatest.vtTOcolorvt(pred))
pred.save("build/" + name + "_z.jpg")
print("accuracy=", np.sum(cm.diagonal()) / (np.sum(cm) + 1))
print(cm)
print("load data")
datatest = segsemdata.makeDFC2015(trainData=False)
datatest = datatest.copyTOcache(outputresolution=50)
nbclasses = len(datatest.setofcolors)
cm = np.zeros((nbclasses, nbclasses), dtype=int)
names = datatest.getnames()
with torch.no_grad():
print("load model")
net = torch.load("build/model.pth")
net = net.to(device)
net.eval()
print("test")
for name in names:
image, label = datatest.getImageAndLabel(name, innumpy=False)
pred = net(image.to(device), datatest.metadata())
