Tianyuan Zhang

Hi, Thanks for your work. I am wondering if there is a full metrics on nuScenes validation split (Tracking). I read your paper, and checked some of the issues, didn't...

Hi, Thanks for your great work. I am wondering if there is a full metrics on nuScenes validation split (Tracking). I read your paper, and checked some of the issues,...

first frame of i2v

2

Hi, Congrats on this great work. I noticed that the first frame from I2V model sometimes is not consistent with the condition image. So I am curious about the definition...

no implmentation of `ps.get_camera_world_position`

3

Hi, thanks for the wonderful project and also building tools like polyscope. ``` def save_render_current_view( args, implicit_func, params, cast_frustum, opts, matcaps, surf_color ): root = ps.get_camera_world_position() look, up, left =...

Hi, I fixed a bug in the post processing step in`NuScenesTrackDataset._format_bbox`. This bug will produce mis-matched tracking indexs, results in a large drop in AMOTA, but leaves the RECALL not...

[BUG] Dynamic shape compile hangs forever

1

### Required prerequisites - [x] I have read the documentation . - [x] I have searched the [Issue Tracker](https://github.com/tile-ai/tilelang/issues) that this hasn't already been reported. (comment there if it has.)...

same matmul kernel has very different results with and without autotune.

1

Hi Tilelang team, I am writing a simple kernel that groups two types of matmul to reduce kernel launching and wave quantization overhead, I just find that: 1. when manually...

autotune fails, but manually set block_M, block_N, block_K would work

4

Hi, tilelang team, thanks for such amazing work. I am writing some simple fused GEMM kernel, by manually selecting the block_M, block_N, block_K, it compiles and works. But when turn...