 Guillermo García
|
d03ce45da5
|
train.py freeze-darknet53 shortened to freeze and action store_true
Traing with freeze: python train.py --freeze
Train without freeze: python train.py
Note: in the actual version freeze is only for first epoche
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
868a116750
|
train.py remove hardcoded weights/ path for weights.
If I want to store my weights in 'weights2' path:
python train.py --weights-path weights2
Default is the same: weights
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
9c0c1f23ab
|
scripts: use data config defined class names
Shorten name of --data-config-path argument to --data-config
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
89daa407e5
|
train.py report argument as store_true
Default is false: python train.py
If want the report: python train.py --report
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
b1fb6fa33d
|
train.py resume argument as store_true
Default is false.
If want to resume, call train.py --resume
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
c807c16b79
|
Fix argument parser bad practice
Keep parsing inside __main__ block and call methods with arguments
Add double -- for long argument names (- reserved for shortcuts)
|
2018-12-05 16:57:16 +01:00 |
|
|
Guillermo García
|
5a566454f5
|
Extract seed and cuda initialization utils
|
2018-12-05 11:55:27 +01:00 |
|