HengYuD

From the log file, the loss_cls does not decreased. Maybe you could check your classes setting.

The model will replace the classifier head and finetune it with N-way K-shot setting while testing. 1 eposide = 5way*(5support+15querys)=100pictures for your setting 5shot/10shot/15shot/20shot, the number of picture is not...

5-shot could test 1-shot model with '../../_base_/meta_test/xxx_meta-test_5way-5shot.py' config for 1shot test the step is 150, so 5shot is 600 just like the config file show meta_finetune_cfg = dict( num_steps=600, optimizer=dict(...

And link https://filterpy.readthedocs.io/en/latest/# is version of 1.4.4 documentation but 1.4.5 is available. Could you provide the examples of CKF?

always showing "Warning: image_path not found for reconstruction" is the warning of pycolmap#493 I tried to convert .pt to .ply with#234 #404 but missed the face vertex_indices in .pt so...

try： pip install --upgrade setuptools wheel

you may used mismatched gaussian_splatting/submodules/diff-gaussian-rasterization the new 3dgs add the depth in rasterizer try pip install the repo submodules or just modify the code: rendered_image, radii, _ = rasterizer(