Closed srbhchandra closed 2 years ago
Truly infinite IAM synthetic paragraphs dataset since every datum in every batch will be different.
Among the best runs so far: https://wandb.ai/fullstackdeeplearning/fsdl-text-recognizer-2022-training/runs/2xybxxqe More importantly, it is 36% faster than the previous case of almost infinite IAM synthetic paragraphs where the paragraphs were randomly recreated every n=20 epochs.
n=20
I didn't make a report specific to this run but made a report combining all the important runs at https://wandb.ai/fullstackdeeplearning/fsdl-text-recognizer-2022-training/reports/Final-set-of-all-worthwhile-runs--VmlldzoyNTM5MjU5.
Signed-off-by: srbhchandra srbhchandra@gmail.com
superseded by #50
Truly infinite IAM synthetic paragraphs dataset since every datum in every batch will be different.
Among the best runs so far: https://wandb.ai/fullstackdeeplearning/fsdl-text-recognizer-2022-training/runs/2xybxxqe More importantly, it is 36% faster than the previous case of almost infinite IAM synthetic paragraphs where the paragraphs were randomly recreated every
n=20
epochs.I didn't make a report specific to this run but made a report combining all the important runs at https://wandb.ai/fullstackdeeplearning/fsdl-text-recognizer-2022-training/reports/Final-set-of-all-worthwhile-runs--VmlldzoyNTM5MjU5.
Signed-off-by: srbhchandra srbhchandra@gmail.com