search

davidsandberg / facenet

Face recognition using Tensorflow
MIT License
13.83k stars 4.81k forks source link

Something went wrong with the epoch number #994

Open g3mini28 opened 5 years ago

g3mini28 commented 5 years ago

Hi,

I am currently trying to finetune the VGGFace2 pretrained model and train using triplet loss. I am training on my own dataset with the following parameters:

python src/train_tripletloss.py --logs_base_dir logs/facenet/TripletLoss_Training --models_base_dir models/facenet/tripletloss --data_dir dataset/own_dataset_160 --image_size 160 --model_def models.inception_resnet_v1 --optimizer RMSPROP --learning_rate 0.01 --weight_decay 1e-4 --pretrained_model Pre-Trained_Model/Inception_ResNet_v1/VGGFace2/20180402-114759/model-20180402-114759.ckpt-275 --embedding_size 512 --images_per_person 10 --people_per_batch 6 --batch_size 12 --epoch_size 100 --max_nrof_epoch 5

I have 10 classes with 10 images each, so a total of 100 images. However, the epoch number jumped from 3 to 5, and after 100 iterations, it did not refresh. Instead continuing till 130+, what went wrong and is there something which I have done wrong? Any help would be much appreciated.

Capture
thelastfunction commented 5 years ago

for finetuning 10 classes are very very less...you should try fine tuning with more data (around 1000+ classes)