machinelp / pytorch_image_classifier

Image classification: efficientnet/resnest/seresnext/.....

Hi 👋, I'm MachineLP

• 🌱 I’m currently learning ...

pytorch_image_classifier's Issues

I don’t know what to do with my own image data set, I hope to get your reply, thank you

Hello, I now want to use resnest to train my own image data set. What I do is image classification, but how to convert my image file into data.csv, because the data processing in the README is a bit unclear, I hope to get Thank you for your reply!

> Hello, I now want to use resnest to train my own image data set. What I do is image classification, but how to convert my image file into data.csv, because the data processing in the README is a bit unclear, I hope to get Thank you for your reply!

Hello, I now want to use resnest to train my own image data set. What I do is image classification, but how to convert my image file into data.csv, because the data processing in the README is a bit unclear, I hope to get Thank you for your reply!

Just convert your data to this format：https://github.com/MachineLP/PyTorch_image_classifier/blob/master/data/data.csv.
"filepath": The path of the image.
"target": The label of the image.
"fold": Not needed.

Originally posted by @MachineLP in #5 (comment)

...

模型使用AUC作为评价指标，只能用于二分类吗？

Does InceptionV4 with RNN on multi-label work?

Does InceptionV4 with RNN on multi-label work?

为什么训练集和测试集是一样的呢

替换为tf官方的mobilenetV2后预测错误

inception_v4_rnn_attention的model里inception_v4替换为mobilenetV2后，训练loss能下降，但预测错误，这是神马情况？

Question about model name

Hi！

Great work done here！
I‘m wondering how the model name combined, such as , what 'ns' in 'tf_efficientnet_b4_ns' means?
other like...
'es' in 'tf_efficientnet_es', 'ap' in ‘tf_efficientnet_b6_ap’, etc.

Really makes me confused...

Thank you!

train default dataset, acc unchange.

Fri Dec 11 22:08:15 2020 Fold 2, Epoch 11, lr: 0.0000849, train loss: 0.03911, valid loss: 0.00720, acc: 47.3684, auc: 0.500000.
Fri Dec 11 22:08:15 2020 Fold 2, Epoch 12
loss: 0.03136, smth: 0.02190: 100%|
100%|
Fri Dec 11 22:08:20 2020 Fold 2, Epoch 12, lr: 0.0000565, train loss: 0.02190, valid loss: 0.00529, acc: 47.3684, auc: 0.500000.
Fri Dec 11 22:08:20 2020 Fold 2, Epoch 13
loss: 0.00997, smth: 0.02012: 100%|
100%|
Fri Dec 11 22:08:24 2020 Fold 2, Epoch 13, lr: 0.0000327, train loss: 0.02012, valid loss: 0.00648, acc: 47.3684, auc: 0.500000.
Fri Dec 11 22:08:24 2020 Fold 2, Epoch 14
loss: 0.06338, smth: 0.05295: 100%|

Fri Dec 11 22:08:29 2020 Fold 2, Epoch 14, lr: 0.0000149, train loss: 0.05295, valid loss: 0.00702, acc: 47.3684, auc: 0.500000.
Fri Dec 11 22:08:29 2020 Fold 2, Epoch 15
loss: 0.00576, smth: 0.01205: 100%|
Fri Dec 11 22:08:35 2020 Fold 2, Epoch 15, lr: 0.0000038, train loss: 0.01205, valid loss: 0.00804, acc: 47.3684, auc: 0.500000.
77 19
Fri Dec 11 22:08:35 2020 Fold 3, Epoch 1
loss: 0.65180, smth: 0.65786: 100%|
Fri Dec 11 22:08:40 2020 Fold 3, Epoch 1, lr: 0.0000300, train loss: 0.65786, valid loss: 0.64426, acc: 47.3684, auc: 0.500000.

模型验证上后些许问题，可否修复一下

ckpt_pb 导出 rnn模型失败

[libprotobuf FATAL google/protobuf/wire_format.cc:830] CHECK failed: (output->ByteCount()) == (expected_endpoint): : Protocol message serialized to a size different from what was originally expected. Perhaps it was modified by another thread during serialization?