MPI-Dortmund / tomotwin-cryoet

cryo-ET particle picking by representation and metric learning
Mozilla Public License 2.0
32 stars 6 forks source link

[Request] save tomotwin result files in float32 #107

Open kimdn opened 1 month ago

kimdn commented 1 month ago

My napari can open my raw/tomotwin input tomogram (024_06_06_10p4_Apix_MED4_TS_14_SIRT_CTF_no_Au_rec.mrc) well as Screenshot 2024-09-23 at 4 05 13 PM

However, when I opened my umap (2024_06_06_10p4_Apix_MED4_TS_14_SIRT_CTF_no_Au_rec_a10_embeddings.tumap) with my napari, I saw

Screenshot 2024-09-23 at 2 11 38 PM

It turns out that some of my x,y,z values are too big to be represented by float16 as Screenshot 2024-09-23 at 2 38 27 PM Screenshot 2024-09-23 at 2 40 17 PM

Indeed, when I resaved them as float32 by df_embeddings[['X', 'Y', 'Z']] = df_embeddings[['X', 'Y', 'Z']].astype('float32') now it didn't give me NaN Screenshot 2024-09-23 at 3 19 02 PM

However, still my napari can't load even this new float32 retyped umap (2024_06_06_10p4_Apix_MED4_TS_14_SIRT_CTF_no_Au_rec_a10_embeddings_float32.tumap) as

Screenshot 2024-09-23 at 3 51 50 PM

thorstenwagner commented 1 month ago

I dont see you coordinate are bigger than 32 767 - why should float16 be the problem