about the reproducible result on S3DIS dataset

[amiltonwong]

Hi, @lightaime ,

According to the provided pretrained models, the result only gives no greater then mIOU=53. However, the result in your paper had mentioned the mIOU performance is 60.0 in Table 2. Could you release the model related to mIOU=60.0?

THX!

[lightaime]

Hi Amiltonwong,

To Clarify the misunderstanding, the mIOU 53 is the performance on Area 5. And mIOU 60.0 is the average mIOU over 6 areas. The original implementation is in the Tensorflow repo. You can get all the pretrained models on 6 areas here https://github.com/lightaime/deep_gcns . As for the pytorch pretrained, we only ran the experiment on Area 5. We may run the rest once we have idle GPUs. Thanks.

On Fri, Dec 27, 2019 at 9:14 AM amiltonwong notifications@github.com wrote:

Hi, @lightaime https://github.com/lightaime ,

According to the provided pretrained models https://github.com/lightaime/deep_gcns_torch/tree/master/examples/sem_seg_dense#pretrained-models, the result only gives no greater then mIOU=53. However, the result in your paper had mentioned the mIOU performance is 60.0 in Table 2. Could you release the model related to mIOU=60.0?

THX!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/lightaime/deep_gcns_torch/issues/13?email_source=notifications&email_token=AFUJTYFUPFBRXOXJ6FCNTZDQ2W2PJA5CNFSM4J7YDHZ2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC2KS4Q, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFUJTYB6YURIXW3NCCEQCI3Q2W2PJANCNFSM4J7YDHZQ .

[a18700]

Hi Amiltonwong, To Clarify the misunderstanding, the mIOU 53 is the performance on Area 5. And mIOU 60.0 is the average mIOU over 6 areas. The original implementation is in the Tensorflow repo. You can get all the pretrained models on 6 areas here https://github.com/lightaime/deep_gcns . As for the pytorch pretrained, we only ran the experiment on Area 5. We may run the rest once we have idle GPUs. Thanks. … On Fri, Dec 27, 2019 at 9:14 AM amiltonwong @.***> wrote: Hi, @lightaime https://github.com/lightaime , According to the provided pretrained models https://github.com/lightaime/deep_gcns_torch/tree/master/examples/sem_seg_dense#pretrained-models, the result only gives no greater then mIOU=53. However, the result in your paper had mentioned the mIOU performance is 60.0 in Table 2. Could you release the model related to mIOU=60.0? THX! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#13?email_source=notifications&email_token=AFUJTYFUPFBRXOXJ6FCNTZDQ2W2PJA5CNFSM4J7YDHZ2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IC2KS4Q>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFUJTYB6YURIXW3NCCEQCI3Q2W2PJANCNFSM4J7YDHZQ .

hello authors,

we want to get overall mIOU(6 areas) for my own trained models, but how can we get it in pytorch?

is it enough for altering from 5 to 1, ..., 6 of below sentence of attaching code, 'test_dataset = GeoData.S3DIS(opt.test_path, 5, False, pre_transform=T.NormalizeScale())' ?

======

def main(): opt = OptInit().initialize()

print('===> Creating dataloader...')
test_dataset = GeoData.S3DIS(opt.test_path, 5, False, pre_transform=T.NormalizeScale())
test_loader = DenseDataLoader(test_dataset, batch_size=opt.batch_size, shuffle=False, num_workers=0)
opt.n_classes = test_loader.dataset.num_classes
if opt.no_clutter:
    opt.n_classes -= 1

======

If the fast reply possible, we will be really grateful of you.

Thank you!