Skip to content

Releases: koshokai/HorseVision

v1.0 — pretrained checkpoints

Choose a tag to compare

@koshokai koshokai released this 23 Sep 15:17

Six checkpoints for the dual-fisheye panoramic depth model (2.2 GB total).

Scene Model Asset
indoor (20 m) fisheye, self-supervised (HM3D) indoor_self_supervised_fish_hm3d_self_fish.tar
indoor fisheye, supervised (Structured3D) indoor_supervised_st3D_sup_fish.tar
indoor MiniDA3 panoramic refiner indoor_supervised_erp_sup_erp.pth
outdoor (80 m) fisheye, self-supervised (CARLA) outdoor_self_supervised_fish_carla_self_fish.tar
outdoor fisheye, supervised (CARLA) outdoor_supervised_fish_carla_sup_fish.tar
outdoor MiniDA3 panoramic refiner outdoor_supervised_erp_sup_erp.pth

Recommended: python download_models.py — places every file at the path the
scripts expect and verifies SHA-256. See models_sha256.txt for the checksums.