I have a tiny problem where the folder names that are selected for class named are confused for the transfer learning tutorial. Could you please guide how to fix it?
*Adding the ipynb here because it could not be loaded in gist even though I uploaded it
My category names are 1,2, … , 16
./test : 135
./test/1 : 5
./test/2 : 2
./test/3 : 4
./test/4 : 63
./test/5 : 2
./test/6 : 0
./test/7 : 12
./test/8 : 8
./test/9 : 13
./test/10 : 3
./test/11 : 1
./test/12 : 3
./test/13 : 3
./test/14 : 6
./test/15 : 2
./test/16 : 8
And the accuracies I get from the tutorial are:
class 1 --> accuracy: 80.00, correct predictions: 4, all: 5
class 2 --> accuracy: 33.33, correct predictions: 1, all: 3
class 3 --> accuracy: 0.00, correct predictions: 0, all: 1
class 4 --> accuracy: 0.00, correct predictions: 0, all: 3
class 5 --> accuracy: 0.00, correct predictions: 0, all: 3
class 6 --> accuracy: 33.33, correct predictions: 2, all: 6
class 7 --> accuracy: 0.00, correct predictions: 0, all: 2
class 8 --> accuracy: 12.50, correct predictions: 1, all: 8
class 9 --> accuracy: 50.00, correct predictions: 1, all: 2
class 10 --> accuracy: 25.00, correct predictions: 1, all: 4
class 11 --> accuracy: 93.65, correct predictions: 59, all: 63
class 12 --> accuracy: 50.00, correct predictions: 1, all: 2
class 13 --> accuracy: nan, correct predictions: 0, all: 0
class 14 --> accuracy: 91.67, correct predictions: 11, all: 12
class 15 --> accuracy: 37.50, correct predictions: 3, all: 8
class 16 --> accuracy: 69.23, correct predictions: 9, all: 13
total correct: 93, total samples: 135
As you see, here class 4 is considered to be 11.
The following is the part of the code used for calculating accuracies:
import ntpath
from torch.utils.data.sampler import WeightedRandomSampler
model_ft.eval()
nb_classes = 16
import torch.nn.functional as F
confusion_matrix = torch.zeros(nb_classes, nb_classes)
_classes = []
_preds = []
predicted_labels = []
class_probs = torch.Tensor()
im_paths = []
with torch.no_grad():
for i, (inputs, classes, im_path) in enumerate(dataloaders['test']):
im_paths.append(im_path)
inputs = inputs.to(device)
classes = classes.to(device)
classes_list = classes.cpu().detach().numpy().tolist()
_classes[:]=[i+1 for i in classes_list]
outputs = model_ft(inputs)
class_probs = class_probs.cuda()
class_probs = torch.cat((class_probs, F.softmax(outputs, 1)))
_, preds = torch.max(outputs, 1)
preds_list = preds.cpu().detach().numpy().tolist()
_preds[:]=[i+1 for i in preds_list]
predicted_labels.append(preds.cpu().detach().numpy().tolist())
for t, p in zip(classes.view(-1), preds.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
print(confusion_matrix)
per_class_accuracies = (confusion_matrix.diag()/confusion_matrix.sum(1)).cpu().detach().numpy().tolist()
print(','.join("{:2.04f}".format(x) for x in per_class_accuracies))
total_correct = 0
total = 0
for i in range(nb_classes):
total_correct += int(confusion_matrix[i][i].numpy())
total += int(confusion_matrix.sum(dim=1)[i].numpy())
print("class {:d} --> accuracy: {:.2f}, correct predictions: {:d}, all: {:d}".format(i+1, (confusion_matrix.diag()/confusion_matrix.sum(1))[i]*100, int(confusion_matrix[i][i].numpy()), int(confusion_matrix.sum(dim=1)[i].numpy())))
print("total correct: {}, total samples: {}".format(total_correct, total))
flattened_im_paths = flattened = [item for sublist in im_paths for item in sublist]
print("length is: ", len(flattened_im_paths))
for i in range(len(flattened_im_paths)):
class_p = class_probs[i].cpu().detach().numpy().tolist()
print('{}, {}'.format(ntpath.basename(flattened_im_paths[i]), class_p))
