Hi, thanks for your great work! Now I am trying to test the trained model (with weight cpjump_cellpaint_bf_channelvit_small_p8_with_hcs_supervised) on jumpcp with my custom dataloader (since I don't want so many config files in my project). The code for data generation is:
def __getitem__(self, index):
if self.well_loc[index] not in self.well2lbl[self.perturbation_type]:
# this well is not labeled
return None
image = self.read_im(self.data_path[index]) #/ 255.
image = self.transform(image)
label = self.well2lbl[self.perturbation_type][self.well_loc[index]]
return {'image': image, 'label': label, 'channels': np.arange(8)}
But the results are not good (~0.2% accuracy). Can you give me some suggestion how to apply the trained model with a custom dataloader (or how to design the dataloader)