Need of prediction mask

[jtyytim]

Dear author,

I read your paper 'Cell Detection with Star-convex Polygons' and I saw your experimental results in Table 1. However, we proposed a method of post-processing that can maybe improve the performance of the experiments based on the predicted results directly. So I would like to ask if you can provide your predicted mask of dataset DSB2018 (in your paper Table 1) so that I can make a comparison based on it.
I am looking forward to hearing from you.

Thiem

[maweigert]

Hi,

you can get instantiate the paper model via

from stardist.models import StarDist2D
model = StarDist2D.from_pretrained('2D_paper_dsb2018')

and then predict the masks yourself.

[jtyytim]

Thanks for your reply. In fact, I got your pretrained model and predicted the mask of 70 images of the dataset DSB2018, but I got AP less than 25% (not 86%) when the threshold is 0.5. Could you please share with me the predicted mask and associated GT that you used in your test part ? [image: image.png] I am looking forward to hearing from you.

…

On Mon, 30 May 2022 at 13:31, Martin Weigert ***@***.***> wrote: Hi, you can get instantiate the paper model via from stardist.models import StarDist2Dmodel = StarDist2D.from_pretrained('2D_paper_dsb2018') and then predict the masks yourself. — Reply to this email directly, view it on GitHub <#199 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AISAN7P2FGR2NMABQQJ7IGDVMSRJPANCNFSM5XKDIRSQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[maweigert]

Did you normalize the images before prediction?

[jtyytim]

In fact, I didn't normalize the images before prediction. But when I tried to normalize the images, I got AP only 1% when the threshold is 0.5.( AP more than 50% when the threshold is 0.1 )

…

On Tue, 31 May 2022 at 15:20, Martin Weigert ***@***.***> wrote: Did you normalize the images before prediction? — Reply to this email directly, view it on GitHub <#199 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AISAN7OZJNG2XMIXAZNJQO3VMYG23ANCNFSM5XKDIRSQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[maweigert]

Did you follow https://github.com/stardist/stardist/blob/master/examples/2D/3_prediction.ipynb ?

[jtyytim]

Thanks for your reply. I followed your link and I used the fonction 'model.predict_instances' instead of 'model.predict'. This time I got a better result: AP is about 60% when the threshold is 0.5, but it's still a little far from your result 86%.

…

On Wed, 1 Jun 2022 at 14:46, Martin Weigert ***@***.***> wrote: Did you follow https://github.com/stardist/stardist/blob/master/examples/2D/3_prediction.ipynb ? — Reply to this email directly, view it on GitHub <#199 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AISAN7OYIZNLRWBK3U526ULVM5LRLANCNFSM5XKDIRSQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[maweigert]

Note that in the paper we used a subset (fluorescence images) of the DSB2018, which you can find here: https://github.com/stardist/stardist/releases/download/0.1.0/dsb2018.zip