OpenDriveLab · sephyli · Oct 29, 2025 · Aug 29, 2025 · Aug 29, 2025 · Aug 29, 2025
diff --git a/.gitignore b/.gitignore
@@ -149,4 +149,10 @@ phoenix*
 debug/
 *projs/
 
-core.*
+core.*
+
+# mmcv package
+mmcv
+
+# vscode config file
+.vscode/
diff --git a/.vscode/launch.json b/.vscode/launch.json
diff --git a/README.md b/README.md
@@ -95,7 +95,7 @@ UniAD is trained in two stages. Pretrained checkpoints of both stages will be re
 
 | Method | Encoder | Tracking<br>AMOTA | Mapping<br>IoU-lane | config | Download |
 | :---: | :---: | :---: | :---: | :---:|:---:| 
-| UniAD-B | R101 | 0.390 | 0.297 |  [base-stage1](projects/configs/stage1_track_map/base_track_map.py) | [base-stage1](https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/uniad_base_track_map.pth) |
+| UniAD-B | R101 | 0.394 | 0.294 |  [base-stage1](projects/configs/stage1_track_map/base_track_map.py) | [base-stage1](https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/uniad_base_track_map.pth) |
 
 
 
@@ -107,7 +107,7 @@ Pre-trained models and results under main metrics are provided below. We refer y
 
 | Method | Encoder | Tracking<br>AMOTA | Mapping<br>IoU-lane | Motion<br>minADE |Occupancy<br>IoU-n. | Planning<br>avg.Col. | config | Download |
 | :---: | :---: | :---: | :---: | :---:|:---:| :---: | :---: | :---: |
-| UniAD-B | R101 | 0.363 | 0.313 | 0.705 | 63.7 | 0.29 |  [base-stage2](projects/configs/stage2_e2e/base_e2e.py) | [base-stage2](https://github.com/OpenDriveLab/UniAD/releases/download/v1.0.1/uniad_base_e2e.pth) |
+| UniAD-B | R101 | 0.380 | 0.314 | 0.794 | 64.0 | 0.29 |  [base-stage2](projects/configs/stage2_e2e/base_e2e.py) | [base-stage2](https://github.com/OpenDriveLab/UniAD/releases/download/v1.0.1/uniad_base_e2e.pth) |
 
 > Planning results on the nuScense benchmark
 
@@ -131,14 +131,14 @@ Pre-trained models and results under main metrics are provided below. We refer y
   <tr>
     <td>UniAD-B</td>
     <td>R101</td>
-    <td>0.48</td>
-    <td>0.96</td>
-    <td>1.65</td>
-    <td>1.03</td>
-    <td>0.05</td>
-    <td>0.17</td>
-    <td>0.71</td>
-    <td>0.31</td>
+    <td>0.29</td>
+    <td>0.89</td>
+    <td>1.53</td>
+    <td>0.90</td>
+    <td>0.15</td>
+    <td>0.20</td>
+    <td>0.53</td>
+    <td>0.29</td>
   </tr>
 </table>
 

diff --git a/docs/DATA_PREP.md b/docs/DATA_PREP.md
@@ -14,12 +14,13 @@ mkdir data
 
 **Prepare UniAD data info**
 
-*Option1: We have already prepared the off-the-shelf data infos for you:*
+*Option1: We have already prepared the off-the-shelf data infos for you in [HuggingFace::OpenDriveLab/UniAD2.0_R101_nuScenes](https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/tree/main/data):*
 ```shell
 cd UniAD/data
 mkdir infos && cd infos
-wget https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/nuscenes_infos_temporal_train.pkl  # train_infos
-wget https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/nuscenes_infos_temporal_val.pkl  # val_infos
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/data/nuscenes_infos_temporal_train.pkl # train_infos
+
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/data/nuscenes_infos_temporal_val.pkl # val_infos
 ```
 
 
@@ -33,10 +34,12 @@ mkdir infos
 ```
 
 **Prepare Motion Anchors**
+
+We already upload motion anchors in [HuggingFace::OpenDriveLab/UniAD2.0_R101_nuScenes](https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/tree/main/data)
 ```shell
 cd UniAD/data
 mkdir others && cd others
-wget https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/motion_anchor_infos_mode6.pkl
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/data/motion_anchor_infos_mode6.pkl
 ```
 
 **The Overall Structure**
@@ -54,10 +57,12 @@ UniAD
 │   ├── nuscenes/
 │   │   ├── can_bus/
 │   │   ├── maps/
+│   │   ├── lidarseg/
 │   │   ├── samples/
 │   │   ├── sweeps/
 │   │   ├── v1.0-test/
 │   │   ├── v1.0-trainval/
+│   │   ├── v1.0-mini/ 
 │   ├── infos/
 │   │   ├── nuscenes_infos_temporal_train.pkl
 │   │   ├── nuscenes_infos_temporal_val.pkl

diff --git a/docs/INSTALL.md b/docs/INSTALL.md
@@ -12,33 +12,14 @@ conda activate uniad2.0
 pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
 ```
 
-**c. GCC: Make sure gcc>=5 in conda env.**
+**c. Install mmcv-series packages.**
 ```shell
-# If gcc is not installed:
-# conda install -c omgarcia gcc-6 # gcc-6.2
-
-export PATH=YOUR_GCC_PATH/bin:$PATH
-# Eg: export PATH=/mnt/gcc-5.4/bin:$PATH
-```
-
-**d. CUDA: Before installing MMCV family, you need to set up the CUDA_HOME (for compiling some operators on the gpu).**
-```shell
-export CUDA_HOME=YOUR_CUDA_PATH/
-# Eg: export CUDA_HOME=/mnt/cuda-11.8/
-```
-
-
-**e. Install mmcv-series packages.**
-```shell
-git clone https://github.com/open-mmlab/mmcv.git & cd mmcv
-git checkout v1.6.0
-export MMCV_WITH_OPS=1 MMCV_CUDA_ARGS=-std=c++17
-pip install -v -e .
+pip install -v mmcv-full==1.6.1 -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.0/index.html
 pip install mmdet==2.26.0 mmsegmentation==0.29.1 mmdet3d==1.0.0rc6
 ```
 
 
-**h. Install UniAD.**
+**d. Install UniAD.**
 ```shell
 cd ~
 git clone https://github.com/OpenDriveLab/UniAD.git
@@ -47,19 +28,23 @@ pip install -r requirements.txt
 ```
 
 
-**i. Prepare pretrained weights.**
+**e. Prepare pretrained weights.**
+
+We release our pretrained weights in [HuggingFace::OpenDriveLab/UniAD2.0_R101_nuScenes](https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/tree/main/ckpts)
+
 ```shell
-mkdir ckpts && cd ckpts
+mkdir ckpts & cd ckpts
+# r101_dcn_fcos3d_pretrain.pth (from bevformer)
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/ckpts/r101_dcn_fcos3d_pretrain.pth
 
-# Pretrained weights of bevformer
-# Also the initial state of training stage1 model
-wget https://github.com/zhiqi-li/storage/releases/download/v1.0/bevformer_r101_dcn_24ep.pth
+# bevformer_r101_dcn_24ep.pth
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/ckpts/bevformer_r101_dcn_24ep.pth
 
-# Pretrained weights of stage1 model (perception part of UniAD)
-wget https://github.com/OpenDriveLab/UniAD/releases/download/v1.0/uniad_base_track_map.pth
+# uniad_base_track_map.pth
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/ckpts/uniad_base_track_map.pth
 
-# Pretrained weights of stage2 model (fully functional UniAD)
-wget https://github.com/OpenDriveLab/UniAD/releases/download/v1.0.1/uniad_base_e2e.pth
+# uniad_base_e2e.pth
+wget https://huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes/resolve/main/ckpts/uniad_base_e2e.pth
 ```
 
 ---

diff --git a/docs/TRAIN_EVAL.md b/docs/TRAIN_EVAL.md
@@ -4,7 +4,7 @@
 Please make sure you have prepared the environment and the nuScenes dataset. You can check it by simply evaluating the pre-trained first-stage(track_map) model as follows:
 ```shell
 cd UniAD
-./tools/uniad_dist_eval.sh ./projects/configs/stage1_track_map/base_track_map.py ./ckpts/uniad_base_track_map.pth 8
+./tools/uniad_dist_eval.sh ./projects/configs/stage1_track_map/base_track_map.py ./ckpts/uniad_base_track_map.pth 8 # (8 is the number of GPUs)
 
 # For slurm users:
 # ./tools/uniad_slurm_eval.sh YOUR_PARTITION ./projects/configs/stage1_track_map/base_track_map.py ./ckpts/uniad_base_track_map.pth 8
@@ -13,9 +13,9 @@ If everything is prepared properly, the output results should be:
 
 ```
 Aggregated results: 
-AMOTA	0.390 
-AMOTP	1.300
-RECALL	0.489
+AMOTA	0.394 
+AMOTP	1.316
+RECALL	0.484
 ```
 
 **Note**: If you evaluate with different number of GPUs rather than 8, the results might be slightly different.
@@ -45,6 +45,15 @@ The first-stage training takes ~ 50 GB GPU memory, ~ 2 days for 6 epochs on 8 A1
 
 The second-stage training takes ~ 17 GB GPU memory, ~ 4 days for 20 epochs on 8 A100 GPUs.
 * **NOTE**: Compared to the first-stage, much less GPU memory is required because we freeze the BEV encoder in this stage to focus on learning task-specific queries. Due to this, you can run the second-stage training on V100 or 3090 devices. -->
+### Bevformer Backbone Training
+we integrated BEVFormer training (you also can use other BEV backbone) into the UniAD code repository in this version.
+```shell
+# Training
+./tools/uniad_dist_train.sh ./projects/configs/bevformer/bevformer_base.py N_GPUS
+
+# Evaluation
+./tools/uniad_dist_train.sh ./projects/configs/bevformer/bevformer_base.py UniAD/ckpts/bevformer_r101_dcn_24ep.pth N_GPUS
+```
 
 
 ### Training Command