JdbermeoUZH MastersThesisUIASegmentation issues

JdbermeoUZH / MastersThesisUIASegmentation

MIT License

0 stars 0 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Refactor forward_seg_pass out of the TTA classes

#108 JdbermeoUZH opened 15 hours ago
0
Refactor: Refactor TTA classes

#107 JdbermeoUZH opened 15 hours ago
0
Train LDDPM with using an image encoder for the segmentation so that we can use it in the cross attention layers

#106 JdbermeoUZH opened 6 days ago
0
Train LDDPM with with image min SNR strategy

#105 JdbermeoUZH opened 6 days ago
0
Experiment: Test if not using drop last or repeat dataset has an effect on the UMC -> NUHS experiments

#104 JdbermeoUZH opened 1 week ago
0
Experiment: Repeat TTA-DAE exp on hcp_t1 on the best performing setting with the DAE trained with the respective dice loss used

#103 JdbermeoUZH opened 1 week ago
0
train our own VAE to improve the current reconstruction error

#102 JdbermeoUZH opened 1 week ago
0
Measure average validation score during training of segmentation models over the entire dataset (instead of the per image averages)

#101 JdbermeoUZH opened 2 weeks ago
0
Finish quantifying reported gap for the different datasets in FoE paper (seems there is not too much gap to go)

#100 JdbermeoUZH opened 2 weeks ago
0
Do BN_Adapt properly instead of updating the running mean and avg fully with that of the new instance

#99 JdbermeoUZH opened 1 month ago
0
Fix how atlas is created, mask for background class is wrongly assigned to 0. So pixels that are always background have pixel distributions that sum to 0

#98 JdbermeoUZH closed 1 month ago
0
Retest Intensity normalization strategy for TTA-DDPM

#97 JdbermeoUZH opened 1 month ago
0
Target Domain Statistics should be a moving estimate. Starting at the source domain is probably not the best bet

#96 JdbermeoUZH opened 1 month ago
0
Carry out TTA exps with new LDDPMs

#95 JdbermeoUZH opened 1 month ago
0
Train LDDPM with different views of dataset

#94 JdbermeoUZH opened 1 month ago
0
Train LDDPM

#93 JdbermeoUZH opened 1 month ago
0
Verify that Neerav's DiceLoss is well defined and behaves properly

#92 JdbermeoUZH opened 1 month ago
0
Create BaseTTA class where the evaluation procedure is implemented

#91 JdbermeoUZH closed 1 week ago
0
Fix visualization of segmentation progress across volume for VU dataset

#90 JdbermeoUZH closed 1 week ago
0
Check if a lower dae update frequency gets rid of the noisyness in the DAE_TTA loop

#89 JdbermeoUZH closed 4 months ago
0
Decide whether to always calculate TD statistics for DDPM

#88 JdbermeoUZH closed 4 months ago
0
Modify my TTA object to measure/log specific labels of interest

#87 JdbermeoUZH closed 1 month ago
0
make computing and logging predictions and metrics with normalizing pixel intensities a standard step whenever calculating statistics for the TD

#86 JdbermeoUZH closed 1 month ago
0
Run cDDPM only experiments with intensity normalization (standarization and min max)

#85 JdbermeoUZH closed 4 months ago
1
Measure DAE baseline with model with stronger DA

#84 JdbermeoUZH closed 5 months ago
1
Add normalization of intensities before evaluating the segmentation network

#83 JdbermeoUZH closed 4 months ago
0
Maybe add a batch normalization (histogram or min/max) that btw. normalization and segmentation network to center and rescale appropriately the intensities under domain shift

#82 JdbermeoUZH closed 1 month ago
0
Verify the better convergence to a solution is not because of the conditioning on multiplication

#81 JdbermeoUZH closed 5 months ago
0
Use lessons learned so far and launch TTA exps for WMH + Synthseg labels

#80 JdbermeoUZH closed 1 month ago
0
Launch TTA experiments for different regularization values of ZNCC

#79 JdbermeoUZH closed 1 month ago
0
Add flag for classifier free guidance weights

#78 JdbermeoUZH closed 5 months ago
0
finish modifying how t gets fed to both conditional and unconditional mode

#77 JdbermeoUZH closed 1 month ago
2
Add measuring the difference in scores between source domain and target domain (the new loss function for PDS)

#76 JdbermeoUZH closed 4 months ago
0
Check how to avoid overfitting with DDPM in the conditional and unconditional mode

#75 JdbermeoUZH closed 5 months ago
0
Add measuring shift with conditional and conditional + unconditional loss

#74 JdbermeoUZH closed 5 months ago
0
Ideas to try later on: Guide based on uncertainty

#73 JdbermeoUZH closed 1 month ago
0
Train DDPM with original lesion and corresponding tissue from Synthseg

#72 JdbermeoUZH closed 1 month ago
1
Keep lesion and original label to train DDPM

#71 JdbermeoUZH closed 1 month ago
0
Test weighing of pixels to train DDPMs

#70 JdbermeoUZH closed 1 month ago
0
Do removal of stuff ablation

#69 JdbermeoUZH closed 1 month ago
0
Fix how regularization with gradients is working

#68 JdbermeoUZH closed 4 months ago
0
Add EMA flag for the gradients of the DDPM and the DAE

#67 JdbermeoUZH closed 4 months ago
0
Measure performance for Abide California with background mask removal

#66 JdbermeoUZH closed 5 months ago
0
Finish measuring performance of WMH + synthseg labels model

#65 JdbermeoUZH closed 1 month ago
0
train both concat and mult conditioned DDPM for more epochs for the hcp_t1w dataset

#64 JdbermeoUZH closed 1 month ago
0
fix logging issues in my normSeg training script

#63 JdbermeoUZH opened 7 months ago
0
Diff tta

#62 JdbermeoUZH closed 7 months ago
0
finally getting some results that make sense"

#61 JdbermeoUZH closed 8 months ago
0
Fix: Include statistics that might be used for normalizing intensities in each fold in the h5 file

#60 JdbermeoUZH closed 1 month ago
0
Segmentation exp: Train segmentation model without backround supression

#59 JdbermeoUZH closed 1 month ago
0