Model update: Update the parameter initialization of FC layer of SE-ResNeXt and add a README #825
Conversation
BigFishMaster
changed the title
BigFishMaster
force-pushed
the
model_update
branch
from
6111972 to
7c3ec94
Compare
The current code can successfully implement the result of SE-ResNeXt-50.
BigFishMaster
force-pushed
the
model_update
branch
from
7c3ec94 to
7103698
Compare
fluid/image_classification/README.md
Outdated
| # prepare directory | ||
| mkdir ILSVRC2012/ | ||
| tar zxf XXX | ||
| tar zxf YYY |
There was a problem hiding this comment.
Better to give the download URL.
| n01440764/n01440764_13602.JPEG 0 | ||
| n01440764/n01440764_13625.JPEG 0 | ||
| ... | ||
| ``` |
There was a problem hiding this comment.
Can these file lists download in somewhere ? like https://github.com/BVLC/caffe/blob/master/data/ilsvrc12/get_ilsvrc_aux.sh
|
|
||
| ``` | ||
| python train.py --num_layers=50 --batch_size=256 --with_mem_opt=True --parallel_exe=True | ||
| ``` |
There was a problem hiding this comment.
It's better to give the convergent curve.
fluid/image_classification/README.md
Outdated
| |- | :-: |:-: | -: | ||
| |SE-ResNeXt-50 | 77.6%/- | 77.71%/93.63% | 77.42%/93.50% | ||
|
|
||
| ## Finetune a model |
There was a problem hiding this comment.
Need to add the usage here to tell the users how to fine-tune a model.
fluid/image_classification/README.md
Outdated
| ``` | ||
| ## Inference | ||
|
|
||
| The inference process is conducted after each training epoch. |
There was a problem hiding this comment.
Need a infer.py to tell the users how to do inference.
| #if global_step % step_each_epoch == 0: | ||
| # print("epoch={0}, global_step={1},decayed_lr={2} \ | ||
| # (step_each_epoch={3})".format( \ | ||
| # epoch,global_step,decayed_lr,step_each_epoch)) |
There was a problem hiding this comment.
Remove the unused code.
| pool = fluid.layers.pool2d( | ||
| input=input, pool_size=0, pool_type='avg', global_pooling=True) | ||
| ### initializer parameter | ||
| #print >> sys.stderr, "pool shape:", pool.shape |
There was a problem hiding this comment.
Remove the unused code.
| param_attr=fluid.param_attr.ParamAttr( | ||
| initializer=fluid.initializer.Uniform(-stdv, | ||
| stdv))) | ||
| #print >> sys.stderr, "squeeze shape:", squeeze.shape |
There was a problem hiding this comment.
Remove the unused code.
|
|
||
|
|
||
| def shortcut(input, ch_out, stride): | ||
| def shortcut_old(input, ch_out, stride): |
There was a problem hiding this comment.
If shortcut_old is not used, please to remove it.
| else: | ||
| drop = pool | ||
| out = fluid.layers.fc(input=drop, size=class_dim, act='softmax') | ||
| #print >> sys.stderr, "drop shape:", drop.shape |
There was a problem hiding this comment.
Remove the unused code.
| import math | ||
|
|
||
|
|
||
| def cosine_decay(learning_rate, step_each_epoch, epochs=120): |
| layers = args.num_layers | ||
| method = train_parallel_exe if args.parallel_exe else train_parallel_do | ||
| init_model = args.init_model if args.init_model else None | ||
| pretrained_model = args.pretrained_model if args.pretrained_model else None |
There was a problem hiding this comment.
上面学习率调整测绿没加cosine_decay
2 participants
fix #826