Comments (7)
kparichay
commented on May 10, 2024
1
Tensorflow VGG evaluation code has some issues:
Below is the log when using tf keras API (fit) to train for the first 50 epochs
78/78 [==============================] - 3s 43ms/step - loss: 4.6548 - acc: 0.0205 - val_loss: 4.6029 - val_acc: 0.0094
Epoch 2/1500
78/78 [==============================] - 1s 15ms/step - loss: 4.4059 - acc: 0.0509 - val_loss: 4.6167 - val_acc: 0.0146
Epoch 3/1500
78/78 [==============================] - 1s 15ms/step - loss: 4.2427 - acc: 0.0768 - val_loss: 4.5593 - val_acc: 0.0271
Epoch 4/1500
78/78 [==============================] - 1s 15ms/step - loss: 4.1213 - acc: 0.0952 - val_loss: 4.5247 - val_acc: 0.0318
Epoch 5/1500
78/78 [==============================] - 1s 15ms/step - loss: 4.0066 - acc: 0.1188 - val_loss: 4.3855 - val_acc: 0.0474
Epoch 6/1500
78/78 [==============================] - 1s 16ms/step - loss: 3.9035 - acc: 0.1377 - val_loss: 4.3042 - val_acc: 0.0490
Epoch 7/1500
78/78 [==============================] - 1s 16ms/step - loss: 3.7998 - acc: 0.1591 - val_loss: 4.1838 - val_acc: 0.0719
Epoch 8/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.6984 - acc: 0.1784 - val_loss: 4.1883 - val_acc: 0.0760
Epoch 9/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.6032 - acc: 0.1982 - val_loss: 4.1549 - val_acc: 0.0812
Epoch 10/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.5146 - acc: 0.2225 - val_loss: 4.1804 - val_acc: 0.0786
Epoch 11/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.4256 - acc: 0.2387 - val_loss: 4.0551 - val_acc: 0.0891
Epoch 12/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.3483 - acc: 0.2549 - val_loss: 3.9600 - val_acc: 0.1068
Epoch 13/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.2747 - acc: 0.2725 - val_loss: 4.0166 - val_acc: 0.1052
Epoch 14/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.2003 - acc: 0.2882 - val_loss: 4.1438 - val_acc: 0.0906
Epoch 15/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.1426 - acc: 0.3012 - val_loss: 4.0813 - val_acc: 0.0938
Epoch 16/1500
78/78 [==============================] - 1s 15ms/step - loss: 3.0739 - acc: 0.3124 - val_loss: 4.1553 - val_acc: 0.0818
Epoch 17/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.9793 - acc: 0.3322 - val_loss: 4.2024 - val_acc: 0.0812
Epoch 18/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.9173 - acc: 0.3499 - val_loss: 4.2972 - val_acc: 0.0880
Epoch 19/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.8376 - acc: 0.3712 - val_loss: 4.2470 - val_acc: 0.0786
Epoch 20/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.7817 - acc: 0.3832 - val_loss: 4.0044 - val_acc: 0.1052
Epoch 21/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.7294 - acc: 0.3922 - val_loss: 3.9743 - val_acc: 0.1052
Epoch 22/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.6682 - acc: 0.4050 - val_loss: 4.4385 - val_acc: 0.0776
Epoch 23/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.5921 - acc: 0.4271 - val_loss: 4.5028 - val_acc: 0.0750
Epoch 24/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.5386 - acc: 0.4356 - val_loss: 4.0264 - val_acc: 0.1214
Epoch 25/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.4445 - acc: 0.4580 - val_loss: 4.1477 - val_acc: 0.0974
Epoch 26/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.3762 - acc: 0.4760 - val_loss: 4.4207 - val_acc: 0.1021
Epoch 27/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.2880 - acc: 0.4990 - val_loss: 4.4048 - val_acc: 0.0922
Epoch 28/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.1901 - acc: 0.5275 - val_loss: 4.3756 - val_acc: 0.0885
Epoch 29/1500
78/78 [==============================] - 1s 15ms/step - loss: 2.1030 - acc: 0.5519 - val_loss: 4.3486 - val_acc: 0.0938
Epoch 30/1500
78/78 [==============================] - 1s 16ms/step - loss: 2.0261 - acc: 0.5718 - val_loss: 4.3443 - val_acc: 0.0938
Epoch 31/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.9430 - acc: 0.5968 - val_loss: 4.2290 - val_acc: 0.0943
Epoch 32/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.8926 - acc: 0.6058 - val_loss: 4.2823 - val_acc: 0.1021
Epoch 33/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.8525 - acc: 0.6130 - val_loss: 4.1456 - val_acc: 0.1151
Epoch 34/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.7682 - acc: 0.6381 - val_loss: 4.1099 - val_acc: 0.1203
Epoch 35/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.6829 - acc: 0.6610 - val_loss: 4.0539 - val_acc: 0.1312
Epoch 36/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.5999 - acc: 0.6825 - val_loss: 4.3446 - val_acc: 0.1099
Epoch 37/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.5149 - acc: 0.7082 - val_loss: 4.3879 - val_acc: 0.1026
Epoch 38/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.4506 - acc: 0.7226 - val_loss: 4.1975 - val_acc: 0.1167
Epoch 39/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.3767 - acc: 0.7432 - val_loss: 4.1234 - val_acc: 0.1193
Epoch 40/1500
78/78 [==============================] - 1s 15ms/step - loss: 1.3143 - acc: 0.7571 - val_loss: 4.3449 - val_acc: 0.1083
Epoch 41/1500
78/78 [==============================] - 1s 15ms/step - loss: 1.2781 - acc: 0.7585 - val_loss: 4.4556 - val_acc: 0.1000
Epoch 42/1500
78/78 [==============================] - 1s 16ms/step - loss: 1.2420 - acc: 0.7735 - val_loss: 4.2794 - val_acc: 0.1219
Epoch 43/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.1939 - acc: 0.7777 - val_loss: 4.2860 - val_acc: 0.1255
Epoch 44/1500
78/78 [==============================] - 1s 18ms/step - loss: 1.1164 - acc: 0.8031 - val_loss: 4.5744 - val_acc: 0.1021
Epoch 45/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.0834 - acc: 0.8104 - val_loss: 4.4412 - val_acc: 0.0984
Epoch 46/1500
78/78 [==============================] - 1s 17ms/step - loss: 1.0332 - acc: 0.8190 - val_loss: 4.4948 - val_acc: 0.1151
Epoch 47/1500
78/78 [==============================] - 1s 17ms/step - loss: 0.9426 - acc: 0.8431 - val_loss: 4.3600 - val_acc: 0.1286
Epoch 48/1500
78/78 [==============================] - 1s 17ms/step - loss: 0.8718 - acc: 0.8648 - val_loss: 4.3338 - val_acc: 0.1271
Epoch 49/1500
78/78 [==============================] - 1s 16ms/step - loss: 0.8247 - acc: 0.8768 - val_loss: 4.4281 - val_acc: 0.1286
Epoch 50/1500
78/78 [==============================] - 1s 16ms/step - loss: 0.8040 - acc: 0.8792 - val_loss: 4.6010 - val_acc: 0.1187Final training accuracy of 87.92 and loss os 0.8040
The same data when used with manual training result is also shown -
#1/1500 - Training Loss: 4.650303 - Training Accuracy: 1.893029 >> [ Accuracy: 0.989583% - Validation Loss : 4.609450 ]
#2/1500 - Training Loss: 4.405456 - Training Accuracy: 5.268429 >> [ Accuracy: 0.989583% - Validation Loss : 4.598562 ]
#3/1500 - Training Loss: 4.242596 - Training Accuracy: 8.082933 >> [ Accuracy: 1.666667% - Validation Loss : 4.592079 ]
#4/1500 - Training Loss: 4.117529 - Training Accuracy: 10.136218 >> [ Accuracy: 1.041667% - Validation Loss : 4.589799 ]
#5/1500 - Training Loss: 4.001147 - Training Accuracy: 12.069311 >> [ Accuracy: 1.250000% - Validation Loss : 4.587155 ]
#6/1500 - Training Loss: 3.889400 - Training Accuracy: 14.132612 >> [ Accuracy: 1.770833% - Validation Loss : 4.585478 ]
#7/1500 - Training Loss: 3.776264 - Training Accuracy: 16.686699 >> [ Accuracy: 1.718750% - Validation Loss : 4.586060 ]
#8/1500 - Training Loss: 3.676095 - Training Accuracy: 19.290865 >> [ Accuracy: 1.979167% - Validation Loss : 4.584372 ]
#9/1500 - Training Loss: 3.573223 - Training Accuracy: 21.424279 >> [ Accuracy: 1.770833% - Validation Loss : 4.583361 ]
#10/1500 - Training Loss: 3.481684 - Training Accuracy: 23.407452 >> [ Accuracy: 1.458333% - Validation Loss : 4.585159 ]
#11/1500 - Training Loss: 3.400188 - Training Accuracy: 25.170272 >> [ Accuracy: 1.666667% - Validation Loss : 4.584146 ]
#12/1500 - Training Loss: 3.327522 - Training Accuracy: 26.221955 >> [ Accuracy: 2.239583% - Validation Loss : 4.583616 ]
#13/1500 - Training Loss: 3.257803 - Training Accuracy: 28.195112 >> [ Accuracy: 2.031250% - Validation Loss : 4.582268 ]
#14/1500 - Training Loss: 3.193364 - Training Accuracy: 29.026442 >> [ Accuracy: 2.552083% - Validation Loss : 4.577587 ]
#15/1500 - Training Loss: 3.110100 - Training Accuracy: 31.330128 >> [ Accuracy: 2.343750% - Validation Loss : 4.579284 ]
#16/1500 - Training Loss: 3.028710 - Training Accuracy: 32.942708 >> [ Accuracy: 2.447917% - Validation Loss : 4.579006 ]
#17/1500 - Training Loss: 2.928692 - Training Accuracy: 35.797276 >> [ Accuracy: 3.020833% - Validation Loss : 4.578534 ]
#18/1500 - Training Loss: 2.850614 - Training Accuracy: 36.919071 >> [ Accuracy: 3.281250% - Validation Loss : 4.577612 ]
#19/1500 - Training Loss: 2.768969 - Training Accuracy: 39.883814 >> [ Accuracy: 2.083333% - Validation Loss : 4.577630 ]
#20/1500 - Training Loss: 2.711938 - Training Accuracy: 40.825321 >> [ Accuracy: 2.135417% - Validation Loss : 4.576182 ]
#21/1500 - Training Loss: 2.643485 - Training Accuracy: 42.377804 >> [ Accuracy: 2.708333% - Validation Loss : 4.573058 ]
#22/1500 - Training Loss: 2.577051 - Training Accuracy: 43.790064 >> [ Accuracy: 2.604167% - Validation Loss : 4.572488 ]
#23/1500 - Training Loss: 2.464163 - Training Accuracy: 46.854968 >> [ Accuracy: 2.500000% - Validation Loss : 4.571971 ]
#24/1500 - Training Loss: 2.365925 - Training Accuracy: 49.188702 >> [ Accuracy: 2.500000% - Validation Loss : 4.571719 ]
#25/1500 - Training Loss: 2.337371 - Training Accuracy: 49.358974 >> [ Accuracy: 2.239583% - Validation Loss : 4.571905 ]
#26/1500 - Training Loss: 2.308354 - Training Accuracy: 50.190304 >> [ Accuracy: 1.979167% - Validation Loss : 4.567413 ]
#27/1500 - Training Loss: 2.257775 - Training Accuracy: 50.701122 >> [ Accuracy: 1.614583% - Validation Loss : 4.566603 ]
#28/1500 - Training Loss: 2.117624 - Training Accuracy: 54.717548 >> [ Accuracy: 1.614583% - Validation Loss : 4.567722 ]
#29/1500 - Training Loss: 2.022456 - Training Accuracy: 57.181490 >> [ Accuracy: 1.510417% - Validation Loss : 4.570506 ]
#30/1500 - Training Loss: 1.981240 - Training Accuracy: 58.213141 >> [ Accuracy: 1.562500% - Validation Loss : 4.572815 ]
#31/1500 - Training Loss: 1.903474 - Training Accuracy: 60.416667 >> [ Accuracy: 1.406250% - Validation Loss : 4.574159 ]
#32/1500 - Training Loss: 1.830212 - Training Accuracy: 62.540064 >> [ Accuracy: 1.510417% - Validation Loss : 4.572250 ]
#33/1500 - Training Loss: 1.764051 - Training Accuracy: 64.132612 >> [ Accuracy: 1.510417% - Validation Loss : 4.576595 ]
#34/1500 - Training Loss: 1.697665 - Training Accuracy: 65.234375 >> [ Accuracy: 1.458333% - Validation Loss : 4.579023 ]
#35/1500 - Training Loss: 1.640916 - Training Accuracy: 66.897035 >> [ Accuracy: 1.354167% - Validation Loss : 4.578002 ]
#36/1500 - Training Loss: 1.579246 - Training Accuracy: 68.399439 >> [ Accuracy: 1.666667% - Validation Loss : 4.578105 ]
#37/1500 - Training Loss: 1.455882 - Training Accuracy: 72.365785 >> [ Accuracy: 1.875000% - Validation Loss : 4.578586 ]
#38/1500 - Training Loss: 1.357298 - Training Accuracy: 74.899840 >> [ Accuracy: 1.666667% - Validation Loss : 4.579042 ]
#39/1500 - Training Loss: 1.279279 - Training Accuracy: 76.913061 >> [ Accuracy: 1.354167% - Validation Loss : 4.581800 ]
#40/1500 - Training Loss: 1.212666 - Training Accuracy: 78.445513 >> [ Accuracy: 1.250000% - Validation Loss : 4.584252 ]
#41/1500 - Training Loss: 1.170250 - Training Accuracy: 79.547276 >> [ Accuracy: 1.093750% - Validation Loss : 4.584248 ]
#42/1500 - Training Loss: 1.149094 - Training Accuracy: 79.557292 >> [ Accuracy: 1.197917% - Validation Loss : 4.582602 ]
#43/1500 - Training Loss: 1.106521 - Training Accuracy: 80.899439 >> [ Accuracy: 1.093750% - Validation Loss : 4.582510 ]
#44/1500 - Training Loss: 1.058206 - Training Accuracy: 81.450321 >> [ Accuracy: 1.145833% - Validation Loss : 4.582915 ]
#45/1500 - Training Loss: 1.037295 - Training Accuracy: 81.430288 >> [ Accuracy: 1.093750% - Validation Loss : 4.585850 ]
#46/1500 - Training Loss: 1.025492 - Training Accuracy: 81.260016 >> [ Accuracy: 1.458333% - Validation Loss : 4.589176 ]
#47/1500 - Training Loss: 1.016521 - Training Accuracy: 80.849359 >> [ Accuracy: 1.666667% - Validation Loss : 4.586395 ]
#48/1500 - Training Loss: 1.006226 - Training Accuracy: 81.320112 >> [ Accuracy: 1.927083% - Validation Loss : 4.587876 ]
#49/1500 - Training Loss: 0.955629 - Training Accuracy: 82.211538 >> [ Accuracy: 1.979167% - Validation Loss : 4.586110 ]
#50/1500 - Training Loss: 0.916124 - Training Accuracy: 83.153045 >> [ Accuracy: 1.666667% - Validation Loss : 4.587056 ]As can be seen, the training result is quite close. However, validation has some bugs (batch normalization mode), which are being fixed. #657 fixes this.
from nntrainer.
taos-ci
commented on May 10, 2024
cibot: Thank you for posting issue #200. The person in charge will reply soon.
from nntrainer.
kparichay
commented on May 10, 2024
- Wait for batch normalization
from nntrainer.
jijoongmoon
commented on May 10, 2024
Close with #639
from nntrainer.
kparichay
commented on May 10, 2024
@jijoongmoon Can you please post the results here from training of both from tensorflow and nntrainer?
from nntrainer.
kparichay
commented on May 10, 2024
Lets keep this open till we post the final comparison.
from nntrainer.
zhoonit
commented on May 10, 2024
I think it is now okay to close this.
from nntrainer.
Related Issues (20)
- Tizen 7.5 Items HOT 1
- Different output between NNTrainer and Tensorflow models after weights initialization HOT 4
- Upgrade Tensorflow 1.x --> Tensorflow 2.x HOT 1
- Support tensor data structure in API HOT 1
- Support free_mem option on inference function even if there is no loss layer. HOT 1
- Need to support the better gradient clipping method. HOT 1
- [graph] Undefined behavior HOT 1
- Print out primary properties when training start HOT 1
- Add Benchmark applications HOT 1
- Support Export & Import Flatbuffer Foramt HOT 2
- Refactoring for several issues HOT 2
- Support swish activation function on LSTM, Attention layers. HOT 1
- Update Android Tensorflow lite version & Flatbuffers version HOT 2
- Errors while building NNTrainer from deb HOT 2
- Errors while building NNTrainer from deb HOT 7
- Cannot create dataset for SimpleShot application HOT 5
- Errors while compiling a file with NNtrainer include HOT 3
- Select ops are not supported in TFlite interpreter HOT 24
- [Pooling] Stride default size issue HOT 1
- Issues while building NNtrainer HOT 5
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from nntrainer.