SamsungLabs / imvoxelnet

[WACV2022] ImVoxelNet: Image to Voxels Projection for Monocular and Multi-View General-Purpose 3D Object Detection
MIT License
283 stars 29 forks source link

no box is drawed on image #46

Closed Calsia closed 2 years ago

Calsia commented 2 years ago

hi,thanks for your code. when i use it to do detection on my own imgs with shape(2304, 1296). there is some box in model's 'output'. but it can't be drawed on img. by the way, i useed the multi_scan line. do you have any suggestion for me? any reponse will help me a lot

Calsia commented 2 years ago

besides, is the extrinsics_mat needed to do normalize? my world's scale maybe is a little big. its unit is mm

filaPro commented 2 years ago

It's hard to help you. Which of our models are you using? You need intrinsic matrix for all models. And you can run our TotalSUNRGBD model without extrinsic matrix. Also you can resize your images to our default 640 x 480 resolution.

Calsia commented 2 years ago

i have tried TotalSUNRGBD, but it also didn't work for me.

i chose the configimvoxelnet/configs/imvoxelnet/imvoxelnet_scannet_fast.py ,and i normalize may extrinsics_mat it into a smale size for the t, i get this output. it seems right . but it still looks like not good.

image