Add Evaluation For Easy of comparison?

[pengwangucla]

Hi Toni, great work,

I find the ref_frame has not been updated for evaluation. Wonder when it will be added ?

Thanks

[pengwangucla]

Another issue is in nerf fusion line 246

You have put gt pose, image & depth in nerf fusion training? wonder whether it is a data leaking?

def fuse(self, data_packets):
        fit = False
        if data_packets:  # data_packets is a dict of data_packets
            for name, packet in data_packets.items():
                if name == "data":
                    fit = self.process_data(packet) # ? add gt pose & image depth in nerf training ? 
                elif name == "slam":
                    fit = self.process_slam(packet)
                    #self.ngp.set_camera_to_training_view(self.ngp.nerf.training.n_images_for_training-1) 
                else:
                    raise NotImplementedError(f"process_{name} not implemented...")
            if fit:
                self.fit_volume()

when name == "data" you also send the packet to nerf training

IC, only (args.fusion == 'nerf' and not slam) case you registerred data

[pengwangucla]

After running the demo. I can't run gui since it is set up on remote server. I just try to render the depth for comparizon.

I render out depth from voi droid-slam output :

But after nerf-fusion with 50000 stop_iter: The renderred depth from self.ngp quality is much worse. I wonder how long we need to train and is there any problem. Or do you plan to publish the correct evaluation so I may check the results ?

Thanks.