File size: 1,421 Bytes
6a176cb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
---
language:
- en
license: apache-2.0
tags:
- object-detection
- 3d-object-detection
- autonomous-driving
- nuscenes
- streampetr
- eva-02
- camera-only
library_name: mmdetection3d
datasets:
- nuscenes
---

# StreamPETR1: StreamPETR with EVA-02 Backbone

This repository contains the source/configuration needed to define the
StreamPETR-style camera-only 3D detector with an EVA-02-Large backbone.

No model weights are included.

## Included

- StreamPETR EVA-02 config:
  `projects/configs/StreamPETR/stream_petr_eva02_flash_800_bs2_seq_24e.py`
- RepDETR3D EVA-02 config:
  `projects/configs/RepDETR3D/repdetr3d_eva02_800_bs2_seq_24e.py`
- Custom model/plugin code under `projects/mmdet3d_plugin/`
- Local MMDetection3D Python/config support files under `mmdetection3d/`
- Utility scripts needed by the configs and data conversion workflow

## Not Included

- Checkpoints or pretrained weights (`.pth`, `.pt`, `.bin`, `.safetensors`, etc.)
- nuScenes data
- Work directories, evaluation outputs, logs, or cache files

## Model Details

- **Architecture:** StreamPETR-style temporal 3D detector
- **Backbone:** EVA-02 Large ViT, patch size 16, embed dim 1024, depth 24, 16 heads
- **Input:** Multi-view camera images
- **Dataset:** nuScenes
- **Task:** Camera-only 3D object detection
- **Framework:** MMDetection3D with a custom StreamPETR plugin

Users must provide their own EVA-02 and detector checkpoints locally.