JaidedAI
Error message appears when training CRAFT with SynthText dataset from scratch?
Hello， I don't know how the size inconsistency occurs when training CRAFT with SynthText dataset from scratch? The training script is: CUDA_VISIBLE_DEVICES=0 python3 trainSynth.py --yaml=syn_train
The information is as follows: 2023-02-08:01:38:12, training_step: 49950|50000, learning rate: 0.00000080, training_loss: 0.03255, avg_batch_time: 29187.57935 2023-02-08:01:38:14, training_step: 49955|50000, learning rate: 0.00000080, training_loss: 0.03671, avg_batch_time: 29189.06155 Error : operands could not be broadcast together with shapes (200,200) (23,0) On generating affinity map, strange box came out. (width: 0, height: 23)
I don't know if anyone can help solve this problem. Thank you.
Hello， I don't know how the size inconsistency occurs when training CRAFT with SynthText dataset from scratch? The training script is: CUDA_VISIBLE_DEVICES=0 python3 trainSynth.py --yaml=syn_train
The information is as follows: 2023-02-08:01:38:12, training_step: 49950|50000, learning rate: 0.00000080, training_loss: 0.03255, avg_batch_time: 29187.57935 2023-02-08:01:38:14, training_step: 49955|50000, learning rate: 0.00000080, training_loss: 0.03671, avg_batch_time: 29189.06155 Error : operands could not be broadcast together with shapes (200,200) (23,0) On generating affinity map, strange box came out. (width: 0, height: 23)
I don't know if anyone can help solve this problem. Thank you.
do you know where the weights file are stored
Firstly, the reasons for this type of error are all from the generation of affinity maps and box generation errors. The specific error is that after processing the map obtained by the model, two boxes are obtained, which are the predicted two chars. After passing through these two boxes, a box is generated, which is the middle part of the two chars used to generate the affinity map. At this point, if the generated box is a single column, but the calculation is based on one row, it will result in its width being zero.
Firstly, the reasons for this type of error are all from the generation of affinity maps and box generation errors. The specific error is that after processing the map obtained by the model, two boxes are obtained, which are the predicted two chars. After passing through these two boxes, a box is generated, which is the middle part of the two chars used to generate the affinity map. At this point, if the generated box is a single column, but the calculation is based on one row, it will result in its width being zero.
How it gonna solve ?
