Wrong l1 and psnr calculation

[unicorne]

For the calculation of l1 and psnr in the training script we have:

        for iteration, (x_val , y_val) in enumerate(data_loader_val): 
        
            real_data = x_val.to(device, non_blocking=True)
            source_data = y_val.to(device, non_blocking=True)
            
            x1_t = torch.cat((torch.randn_like(real_data),source_data),axis=1)
            #diffusion steps
            fake_sample1 = sample_from_model(pos_coeff, gen_diffusive_1, args.num_timesteps, x1_t, T, args)            
            fake_sample1 = to_range_0_1(fake_sample1) ; fake_sample1 = fake_sample1/fake_sample1.mean()
            real_data = to_range_0_1(real_data) ; real_data = real_data/real_data.mean()

            fake_sample1=fake_sample1.cpu().numpy()
            real_data=real_data.cpu().numpy()
            val_l1_loss[0,epoch,iteration]=abs(fake_sample1 -real_data).mean()
            
            val_psnr_values[0,epoch, iteration] = psnr(real_data,fake_sample1, data_range=real_data.max())

        for iteration, (y_val , x_val) in enumerate(data_loader_val): 
        
            real_data = x_val.to(device, non_blocking=True)
            source_data = y_val.to(device, non_blocking=True)
            
            x1_t = torch.cat((torch.randn_like(real_data),source_data),axis=1)
            #diffusion steps
            fake_sample1 = sample_from_model(pos_coeff, gen_diffusive_1, args.num_timesteps, x1_t, T, args)

            
            fake_sample1 = to_range_0_1(fake_sample1) ; fake_sample1 = fake_sample1/fake_sample1.mean()
            real_data = to_range_0_1(real_data) ; real_data = real_data/real_data.mean()
            
            fake_sample1=fake_sample1.cpu().numpy()
            real_data=real_data.cpu().numpy()
            val_l1_loss[1,epoch,iteration]=abs(fake_sample1 -real_data).mean()
            
            val_psnr_values[1,epoch, iteration] = psnr(real_data,fake_sample1, data_range=real_data.max())

You swap x_val and y_val for both loops but both take gen_diffusive_1. The second loop creates a T2 (contrast2) given T1 (contrast1). I think this needs the gen_diffusive_2.