reproduction problem #27
Comments
|
It seems totally failed. I did not meet this problem before. Did you load all pre-trained weights as instructed? |
yes, i load the pre-trained weights and besides i test the best model on pitts 30k that you provide (https://drive.google.com/drive/folders/1FLjxFhKRO-YJQ6FI-DcCMMHDL2K_Hsof) , the score is : |
|
I cannot locate your issue. Maybe provide your full testing log could help. |
this is the log of first evaluation of my initial model:Args:Namespace(arch='vgg16', cache_size=1000, data_dir='/data/yangling/datasets', dataset='pitts', deterministic=False, epochs=5, eval_step=1, features=4096, generations=4, gpu=0, height=480, init_dir='/data/yangling/openIBL/examples/../logs', iters=0, launcher='pytorch', layers='conv5', logs_dir='logs/netVLAD/pitts30k-vgg16/conv5-sare_ind-lr0.001-tuple1-SFRS', loss_type='sare_ind', lr=0.001, margin=0.1, momentum=0.9, neg_num=10, neg_pool=1000, ngpus_per_node=1, nowhiten=False, num_clusters=64, pos_num=10, pos_pool=20, print_freq=10, rank=0, resume='', scale='30k', seed=43, soft_weight=0.5, step_size=5, sync_gather=False, syncbn=True, tcp_port='19294', temperature=[0.07, 0.07, 0.06, 0.05], test_batch_size=8, tuple_size=1, weight_decay=0.001, width=640, workers=2, world_size=1)Pittsburgh dataset loaded
|
if you have retested the model rightly, could you please zip your local folder and send it to my email ([email protected])? I would appreciate it. |
|
It shows that resume='', indicating that you did not load any model weights. |
this is my training log, so resume='' |
And this is my testing log, the model is downloaded from your provided url: Use GPU: 0 for testing, rank no.0 of world_size 1Args:Namespace(arch='vgg16', data_dir='/data/yangling/test/OpenIBL-0.1.0-beta/examples/data', dataset='pitts', features=4096, gpu=0, height=480, lambda_value=0, launcher='pytorch', logs_dir='/data/yangling/test/OpenIBL-0.1.0-beta/examples/logs', ngpus_per_node=1, nowhiten=False, num_clusters=64, print_freq=10, rank=0, reduction=True, rerank=False, resume='logs/ori/model_best.pth.tar', rr_topk=25, scale='30k', sync_gather=False, tcp_port='5017', test_batch_size=6, vlad=True, width=640, workers=2, world_size=1)Pittsburgh dataset loaded
|
I think whether some default settings about dataset preprocessing are wrong with your current code, because the evaluation results of the models you provide and i train are similarly much lower than you report. |
I just run the code without any change and find the initial recall scores is as follow:
Recall Scores:
top-1 1.2%
top-5 4.6%
top-10 8.7%
then I continue the training process util the generation1 epoch2 with the recall score:
Recall Scores:
top-1 1.4%
top-5 6.3%
top-10 11.9%
it seems no obvious promotion, is that right?
The text was updated successfully, but these errors were encountered: