Toy model for coke/pepsi classification
This is a toy model using tensorflow 2.0 for soft drink classification between coke and pepsi cans and bottles, meant for demonstrative purposes. Images were collected from Google image search.
model.py contains the model definition
train.py is the main training file. Hyperparameters are hardcoded.
convert_tflite.py runs a script which converts the saved model to a tflite file
Because only ~70 images are used, all training data is included in the repo. The trained model, in saved_model format, and in tflite format, are also included in this repo.
Training log
Below is a training log for 50 epochs:
Epoch 1, Loss: 0.7484723925590515, Accuracy: 60.000003814697266, Test Loss: 0.7023062109947205, Test Accuracy: 53.333335876464844
Epoch 2, Loss: 0.7400655746459961, Accuracy: 53.333335876464844, Test Loss: 0.6836736798286438, Test Accuracy: 53.333335876464844
Epoch 3, Loss: 0.6902545094490051, Accuracy: 48.88888931274414, Test Loss: 0.6944419741630554, Test Accuracy: 46.66666793823242
Epoch 4, Loss: 0.6925384402275085, Accuracy: 48.88888931274414, Test Loss: 0.6771779656410217, Test Accuracy: 60.000003814697266
Epoch 5, Loss: 0.6740819811820984, Accuracy: 55.55555725097656, Test Loss: 0.656101644039154, Test Accuracy: 53.333335876464844
Epoch 6, Loss: 0.6563219428062439, Accuracy: 66.66667175292969, Test Loss: 0.6177938580513, Test Accuracy: 60.000003814697266
Epoch 7, Loss: 0.3819253146648407, Accuracy: 84.44444274902344, Test Loss: 0.28830215334892273, Test Accuracy: 93.33333587646484
Epoch 8, Loss: 0.4367862939834595, Accuracy: 82.22222137451172, Test Loss: 0.44751861691474915, Test Accuracy: 80.0
Epoch 9, Loss: 0.4134106934070587, Accuracy: 88.8888931274414, Test Loss: 0.26702770590782166, Test Accuracy: 93.33333587646484
Epoch 10, Loss: 0.2825040817260742, Accuracy: 93.33333587646484, Test Loss: 0.22878111898899078, Test Accuracy: 100.0
Epoch 11, Loss: 0.1658862829208374, Accuracy: 97.77777862548828, Test Loss: 0.154488667845726, Test Accuracy: 93.33333587646484
Epoch 12, Loss: 0.24487002193927765, Accuracy: 93.33333587646484, Test Loss: 0.3519488275051117, Test Accuracy: 86.66666412353516
Epoch 13, Loss: 0.25000637769699097, Accuracy: 84.44444274902344, Test Loss: 0.47134777903556824, Test Accuracy: 80.0
Epoch 14, Loss: 0.20285126566886902, Accuracy: 91.11111450195312, Test Loss: 0.13951818645000458, Test Accuracy: 100.0
Epoch 15, Loss: 0.19104258716106415, Accuracy: 97.77777862548828, Test Loss: 0.16895979642868042, Test Accuracy: 93.33333587646484
Epoch 16, Loss: 0.05736321955919266, Accuracy: 100.0, Test Loss: 0.28337863087654114, Test Accuracy: 93.33333587646484
Epoch 17, Loss: 0.15777914226055145, Accuracy: 93.33333587646484, Test Loss: 0.09994355589151382, Test Accuracy: 93.33333587646484
Epoch 18, Loss: 0.03653470799326897, Accuracy: 97.77777862548828, Test Loss: 0.0982230007648468, Test Accuracy: 93.33333587646484
Epoch 19, Loss: 0.09954585880041122, Accuracy: 93.33333587646484, Test Loss: 0.02471625804901123, Test Accuracy: 100.0
Epoch 20, Loss: 0.06976577639579773, Accuracy: 95.55555725097656, Test Loss: 0.16978169977664948, Test Accuracy: 93.33333587646484
Epoch 21, Loss: 0.0688447579741478, Accuracy: 95.55555725097656, Test Loss: 0.09541510790586472, Test Accuracy: 100.0
Epoch 22, Loss: 0.16378824412822723, Accuracy: 91.11111450195312, Test Loss: 0.012079094536602497, Test Accuracy: 100.0
Epoch 23, Loss: 0.0491531565785408, Accuracy: 100.0, Test Loss: 0.08242759108543396, Test Accuracy: 100.0
Epoch 24, Loss: 0.07256065309047699, Accuracy: 93.33333587646484, Test Loss: 0.014410089701414108, Test Accuracy: 100.0
Epoch 25, Loss: 0.054267726838588715, Accuracy: 97.77777862548828, Test Loss: 0.014437764883041382, Test Accuracy: 100.0
Epoch 26, Loss: 0.014979813247919083, Accuracy: 100.0, Test Loss: 0.009632942266762257, Test Accuracy: 100.0
Epoch 27, Loss: 0.035872772336006165, Accuracy: 97.77777862548828, Test Loss: 0.005337590351700783, Test Accuracy: 100.0
Epoch 28, Loss: 0.08602884411811829, Accuracy: 97.77777862548828, Test Loss: 0.004631286486983299, Test Accuracy: 100.0
Epoch 29, Loss: 0.04752058908343315, Accuracy: 97.77777862548828, Test Loss: 0.3451300859451294, Test Accuracy: 86.66666412353516
Epoch 30, Loss: 0.10340915620326996, Accuracy: 95.55555725097656, Test Loss: 0.04207894578576088, Test Accuracy: 100.0
Epoch 31, Loss: 0.05640830472111702, Accuracy: 100.0, Test Loss: 0.03375643491744995, Test Accuracy: 100.0
Epoch 32, Loss: 0.0075826495885849, Accuracy: 100.0, Test Loss: 0.0023693253751844168, Test Accuracy: 100.0
Epoch 33, Loss: 0.003913493826985359, Accuracy: 100.0, Test Loss: 0.07982256263494492, Test Accuracy: 93.33333587646484
Epoch 34, Loss: 0.0014794141752645373, Accuracy: 100.0, Test Loss: 0.04896518215537071, Test Accuracy: 93.33333587646484
Epoch 35, Loss: 0.0031059186439961195, Accuracy: 100.0, Test Loss: 0.00011615739640546963, Test Accuracy: 100.0
Epoch 36, Loss: 0.00037152308505028486, Accuracy: 100.0, Test Loss: 0.14112089574337006, Test Accuracy: 93.33333587646484
Epoch 37, Loss: 0.0011328798718750477, Accuracy: 100.0, Test Loss: 0.14973342418670654, Test Accuracy: 93.33333587646484
Epoch 38, Loss: 0.00011892782640643418, Accuracy: 100.0, Test Loss: 0.23382961750030518, Test Accuracy: 93.33333587646484
Epoch 39, Loss: 0.0011291601695120335, Accuracy: 100.0, Test Loss: 0.0011561453575268388, Test Accuracy: 100.0
Epoch 40, Loss: 0.0015463603194803, Accuracy: 100.0, Test Loss: 0.04765807464718819, Test Accuracy: 100.0
Epoch 41, Loss: 0.0007707966724410653, Accuracy: 100.0, Test Loss: 0.06439944356679916, Test Accuracy: 93.33333587646484
Epoch 42, Loss: 0.001605882658623159, Accuracy: 100.0, Test Loss: 0.16670644283294678, Test Accuracy: 93.33333587646484
Epoch 43, Loss: 0.0012983831111341715, Accuracy: 100.0, Test Loss: 0.000518298358656466, Test Accuracy: 100.0
Epoch 44, Loss: 0.00048058066749945283, Accuracy: 100.0, Test Loss: 0.08776756376028061, Test Accuracy: 93.33333587646484
Epoch 45, Loss: 0.0002973123046103865, Accuracy: 100.0, Test Loss: 0.0013820636086165905, Test Accuracy: 100.0
Epoch 46, Loss: 0.0006491061067208648, Accuracy: 100.0, Test Loss: 0.048505742102861404, Test Accuracy: 93.33333587646484
Epoch 47, Loss: 8.846184937283397e-05, Accuracy: 100.0, Test Loss: 0.05628155171871185, Test Accuracy: 93.33333587646484
Epoch 48, Loss: 0.0008474474889226258, Accuracy: 100.0, Test Loss: 0.032403670251369476, Test Accuracy: 100.0
Epoch 49, Loss: 0.0015330800088122487, Accuracy: 100.0, Test Loss: 0.016267243772745132, Test Accuracy: 100.0
Epoch 50, Loss: 1.5821033230167814e-05, Accuracy: 100.0, Test Loss: 0.49899783730506897, Test Accuracy: 93.33333587646484