update README

Author: liming-ai

README.md

@@ -102,9 +102,10 @@ Please download the model weights and put them into each subset of `checkpoints`
 |:----------:|:------------------------------------------------------------------------------------|
 | LineArt | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/lineart) |
 | Depth | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/depth) |
-| Segmentation | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/seg) |
 | Hed (SoftEdge) | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/hed) |
 | Canny | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/canny) |
+| Segmentation (ADE20K) | [UperNet-R50](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/ade20k_reward-model-UperNet-R50/checkpoint-5000/controlnet), [FCN-R101](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/ade20k_reward-model-FCN-R101-d8/checkpoint-5000/controlnet) |
+| Segmentation (COCOStuff) | [model](https://huggingface.co/limingcv/reward_controlnet/tree/main/checkpoints/cocostuff/reward_5k) |
 
 Please make sure the folder directory is consistent with the test script, then you can eval each model by:
 ```bash
@@ -116,7 +117,7 @@ bash eval/eval_hed.sh
 bash eval/eval_linedrawing.sh
 ```
 
-The ADE20K segmentation results in the arxiv version were tested using images and labels saved in `.jpg` format, which led to errors. We will retest and report the results using images and labels saved in `.png` format.
+The ADE20K segmentation results in the arXiv v1 version of the paper were tested using images and labels saved in `.jpg` format, which resulted in errors. We retested and reported the results using images and labels saved in `.png` format, please refer to our latest arXiv and ECCV Camera Ready releases.
 
 ## 🕹️ Inference
 Please refer to the [Inference Branch](https://github.com/liming-ai/ControlNet_Plus_Plus/tree/inference) or try our [online Huggingface demo](https://huggingface.co/spaces/limingcv/ControlNet-Plus-Plus)

eval/eval_cocostuff.sh

@@ -1,5 +1,6 @@
 # Path to the controlnet weight (can be huggingface or local path)
-export CONTROLNET_DIR="lllyasviel/control_v11p_sd15_seg"
+# export CONTROLNET_DIR="lllyasviel/control_v11p_sd15_seg" # Eval ControlNet
+export CONTROLNET_DIR="checkpoints/cocostuff/reward_5k" # Eval our ControlNet++
 # How many GPUs and processes you want to use for evaluation.
 export NUM_GPUS=8
 # Guidance scale and inference steps
@@ -17,7 +18,7 @@ export DATA_DIR="work_dirs/eval_dirs/Captioned_COCOStuff/validation/${CONTROLNET
 # Evaluation with mmseg api
 mim test mmseg mmlab/mmseg/deeplabv3_r101-d8_4xb4-320k_coco-stuff164k-512x512.py \
  --checkpoint https://download.openmmlab.com/mmsegmentation/v0.5/deeplabv3/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k/deeplabv3_r101-d8_512x512_4x4_320k_coco-stuff164k_20210709_155402-3cbca14d.pth \
- --gpus 4 \
+ --gpus 8 \
  --launcher pytorch \
  --cfg-options test_dataloader.dataset.datasets.0.data_root="${DATA_DIR}" \
  test_dataloader.dataset.datasets.1.data_root="${DATA_DIR}" \