search

smhassanerfani / atlantis

A Benchmark for Semantic Segmentation of Waterbody Images
38 stars 11 forks source link

Segmentation mask image is three color channel #14

Closed prajachintya closed 1 year ago

prajachintya commented 2 years ago

Hi,

I've tried to export images to Segmentation Mask in CVAT, but it came out as three color channel image. How can I get the single-channel mask images?

Thank you :)

smhassanerfani commented 2 years ago

Hi, Unfortunately, CVAT doesn't give you ID maps. I suggest using color2id.py. Just config the labels_info.json based on your own labels list.

Good luck!

prajachintya commented 2 years ago

Thank you for the reply. This algorithm works fine. Btw how did you do the evaluation for the prediction results?

smhassanerfani commented 2 years ago

Hi, Sorry I was a little busy this week, I'll add argparse package to color2id.py to be easier to work with it. Regarding your second question, you can use compute_iou.py. Again, please configure the labels_info.json for this code.

Good Luck!

prajachintya commented 1 year ago

Thank you!

prajachintya commented 1 year ago

No worries; thanks for the update!

By the way, for the AquaNet code, I'm wondering why the number of classes is subtracted by 17?

[image: image.png]

Thank you!

Best Regards, Praja

On Wed, 23 Nov 2022 at 22:16, Mohammad Erfani @.***> wrote:

Hi, Sorry I was a little busy this week, I'll add argparse package to color2id.py to be easier to work with it. Regarding your second question, you can use compute_iou.py https://github.com/smhassanerfani/atlantis/blob/master/utils/compute_iou.py. Again, please configure the labels_info.json https://github.com/smhassanerfani/atlantis/blob/master/utils/labels_info.json for this code.

Good Luck!

— Reply to this email directly, view it on GitHub https://github.com/smhassanerfani/atlantis/issues/14#issuecomment-1325049736, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKOXPBZYNWATIP4MDXLN7DTWJYKJBANCNFSM6AAAAAASF5CIUQ . You are receiving this because you authored the thread.Message ID: @.***>

smhassanerfani commented 1 year ago

According to the manuscript of the paper, "The aquatic path is to segment different types of waterbodies including sea, river, waterfall, wetland and etc., while the non-aquatic path is to segment other categories such as ship and bridge." The manuscript explains further that "Aquatic categories include 17 labels showing just water content in different forms and bodies, e.g., sea, river, lake, etc."

Hope it helps. Mohammad

prajachintya commented 1 year ago

Thank you for your explanation!

I applied PSPNet and AquaNet to 7 classes dataset. One class is for water and the rest 6 classes are non water classes. However the results look like they were reversed.

Is self indices order on AquaNet shows specific purpose? I wondering why the self indices shows are not in order.

Have a great day!

On Wed, Jan 4, 2023 at 11:45 PM Mohammad Erfani @.***> wrote:

According to the manuscript of the paper, "The aquatic path is to segment different types of waterbodies including sea, river, waterfall, wetland and etc., while the non-aquatic path is to segment other categories such as ship and bridge." The manuscript explains further that "Aquatic categories include 17 labels showing just water content in different forms and bodies, e.g., sea, river, lake, etc."

Hope it helps. Mohammad

— Reply to this email directly, view it on GitHub https://github.com/smhassanerfani/atlantis/issues/14#issuecomment-1371019371, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKOXPB35G7GAHESHP7PTSZTWQWEJ7ANCNFSM6AAAAAASF5CIUQ . You are receiving this because you modified the open/close state.Message ID: @.***>

-- Best Regards, Ni Putu Praja Chintya Geodetic Engineering Student Faculty of Engineering Universitas Gadjah Mada Email : @.*** Phone : 08563897789 <085729809494>/082140623448

smhassanerfani commented 1 year ago

Yes! AQUANet is customized for ATLANTIS. I would modify the self.indices based on the new dataset.