fake comparison to see what I changed between MNIS-115 and MNIS-64#60

Open

dalessioluca wants to merge 25 commits intocompare_nv_12from mnist

Commits on Nov 10, 2020

run with noisy_sampling=False at test time
dalessioluca
committed

Commits on Nov 11, 2020

switch nms off after a certain number of epoch
dalessioluca
committed
nms is always on, but threshold value at train and test time are different
dalessioluca
committed
made MNIST dataset with GRID
dalessioluca
committed
dataset created double tensor instead of float
dalessioluca
committed
draw the background when I do generation
dalessioluca
committed
print every 10 epoch
dalessioluca
committed

Commits on Nov 12, 2020

added print statements
dalessioluca
committed
debugging
dalessioluca
committed
removed print statement. Empty solution happens if prob_corr_factor <= 0.5 and noisy_sampling=False. I have fixed by changing prob_corr_factor to 0.55
dalessioluca
committed
added comment and change plotting so that p_map plots 16 inputs images not 8
dalessioluca
committed
visualize reconstruction with and without noisy_sampling
dalessioluca
committed
detached the torch.sign from the g_geco. Added comment
dalessioluca
committed
clean up GECO code
dalessioluca
committed
revert back to OLD geco formulation
dalessioluca
committed
write N_cell constraint based on c_map_before_nms so that all peaks disappear
dalessioluca
committed
g_cell knows about c_map_after_nms. and prob_corr_factor is lower so that p need to start learning from the beginning.
dalessioluca
committed
reverted back to new geco
dalessioluca
committed
fixed typo detahc -> detach
dalessioluca
committed

Commits on Nov 13, 2020

done postprocessing on multi_MNIST
dalessioluca
committed

Commits on Nov 14, 2020

ready to rerun smFISH
dalessioluca
committed
modified ML_params.json to match new code
dalessioluca
committed
comment plt.axis(off) in show_batch
dalessioluca
committed
larger reference images
dalessioluca
committed
done postprocessing on smFISH
DeepLearning VM
committed