Update README.md

README.md

@@ -1,21 +1,23 @@
-# ApDepth: Aiming for Precise Monocular Depth Estimation Based on Diffusion Models
+# SCFDepth: A Single-Step Coarse-to-Fine Diffusion Framework for Monocular Depth Estimation
 
 This repository is based on [Marigold](https://marigoldmonodepth.github.io), CVPR 2024 Best Paper: [**Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation**](https://arxiv.org/abs/2312.02145)
 
 [![Website](doc/badges/badge-website.svg)](https://haruko386.github.io/research)
 [![License](https://img.shields.io/badge/License-Apache--2.0-929292)](https://www.apache.org/licenses/LICENSE-2.0)
-[![Static Badge](https://img.shields.io/badge/build-Haruko386-brightgreen?style=flat&logo=steam&logoColor=white&logoSize=auto&label=steam&labelColor=black&color=gray&cacheSeconds=3600)](https://steamcommunity.com/profiles/76561198217881431/)
 [![Hugging Face Model](https://img.shields.io/badge/🤗%20Hugging%20Face-Model-green)](https://huggingface.co/developy/ApDepth)
 [![Hugging Face Demo](https://img.shields.io/badge/🤗%20Hugging%20Face-Demo-purple)](https://huggingface.co/spaces/developy/ApDepth)
 
-[Haruko386](https://haruko386.github.io/),
-[Shuai Yuan](https://syjz.teacher.360eol.com/teacherBasic/preview?teacherId=23776)
+[**Haruko386**](https://haruko386.github.io/),
+[Shuai Yuan](https://syjz.teacher.360eol.com/teacherBasic/preview?teacherId=23776),
+[Mingbo Lei](https://github.com/Ltohka)
+[Yibo Chen](#)
 
 ![cover](doc/cover.jpg)
 
->We present **ApDepth**, a diffusion model, and associated fine-tuning protocol for monocular depth estimation. Based on Marigold. Its core innovation lies in addressing the deficiency of diffusion models in feature representation capability. Our model followed Marigold, derived from Stable Diffusion and fine-tuned with synthetic data: Hypersim and VKitti, achieved ideal results in object edge refinement.
+>We present **SCFDepth**, a diffusion model, and associated fine-tuning protocol for monocular depth estimation. Based on Marigold. Its core innovation lies in addressing the deficiency of diffusion models in feature representation capability. Our model followed Marigold, derived from Stable Diffusion and fine-tuned with synthetic data: Hypersim and VKitti, achieved ideal results in object edge refinement.
 
 ## 📢 News
+- 2025-10-25: Inspired by DepthMaster, we propose a two-stage loss function training strategy based on `Apepth V1-0`. In the first stage, we perform foundational training using MSE loss. In the second stage, we learn edge structures through FFT loss. Based on this, we introduce Apepth V1-1.
 - 2025-10-09: We propose a novel diffusion-based deep estimation framework guided by pre-trained models.
 - 2025-09-23: We change Marigold from **Stochastic multi-step generation** to **Deterministic one-step perception**
 - 2025-08-10: Trying to make some optimizations in Feature Expression<br>
@@ -23,7 +25,7 @@ This repository is based on [Marigold](https://marigoldmonodepth.github.io), CVP
 
 ## 🚀 Usage
 
-**We offer several ways to interact with Marigold**:
+**We offer several ways to interact with SCFDepth**:
 
 1. A free online interactive demo is available here: <a href="https://huggingface.co/spaces/developy/ApDepth"><img src="https://img.shields.io/badge/🤗%20Hugging%20Face-Demo-purple" height="18"></a>
 
@@ -32,10 +34,13 @@ This repository is based on [Marigold](https://marigoldmonodepth.github.io), CVP
 3. Local development instructions with this codebase are given below.
 
 ## 🛠️ Setup
+The Model was trained on:
+
+- Ubuntu 22.04 LTS, Python 3.12.9,  CUDA 11.8, `NVIDIA RTX 6000 Ada Generation`
 
 The inference code was tested on:
 
-- Ubuntu 22.04 LTS, Python 3.12.9,  CUDA 11.8, GeForce RTX 4090 & GeForce RTX 5080 (pip)
+- Ubuntu 22.04 LTS, Python 3.12.9,  CUDA 11.8, `NVIDIA GeForce RTX 4090`
 
 ### 🪧 A Note for Windows users
 
@@ -50,8 +55,8 @@ We recommend running the code in WSL2:
 Clone the repository (requires git):
 
 ```bash
-git clone https://github.com/Haruko386/ApDepth.git
-cd ApDepth
+git clone https://github.com/Dimon0000000/SCFDepth.git
+cd SCFDepth
 ```
 
 ### 💻 Dependencies
@@ -59,12 +64,18 @@ cd ApDepth
  **Using Conda:** 
     Alternatively, create a Python native virtual environment and install dependencies into it:
 
-    conda create -n apdepth python==3.12.9
-    conda activate apdepth
-    pip install -r requirements.txt
+```bash
+conda create -n SCFDepth python==3.12.9
+conda activate SCFDepth
+pip install -r requirements.txt
+```
+
+> [!NOTE]
+>
+> Keep the environment activated before running the inference script. 
+> Activate the environment again after restarting the terminal session.
+
 
-Keep the environment activated before running the inference script. 
-Activate the environment again after restarting the terminal session.
 
 ## 🏃 Testing on your images
 
@@ -81,8 +92,7 @@ This setting corresponds to our paper. For academic comparison, please run with
 ```bash
 python run.py \
     --checkpoint prs-eth/marigold-v1-0 \
-    --denoise_steps 50 \
-    --ensemble_size 10 \
+    --ensemble_size 1 \
     --input_rgb_dir input/in-the-wild_example \
     --output_dir output/in-the-wild_example
 ```
@@ -94,10 +104,9 @@ You can find all results in `output/in-the-wild_example`. Enjoy!
 The default settings are optimized for the best result. However, the behavior of the code can be customized:
 
 - Trade-offs between the **accuracy** and **speed** (for both options, larger values result in better accuracy at the cost of slower inference.)
-  - `--ensemble_size`: Number of inference passes in the ensemble. For LCM `ensemble_size` is more important than `denoise_steps`. Default: ~~10~~ 5 (for LCM).
-  - `--denoise_steps`: Number of denoising steps of each inference pass. For the original (DDIM) version, it's recommended to use 10-50 steps, while for LCM 1-4 steps. When unassigned (`None`), will read default setting from model config. Default: ~~10 4 (for LCM)~~ `None`.
+  - `--ensemble_size`: Number of inference passes in the ensemble. 
 
-- By default, the inference script resizes input images to the *processing resolution*, and then resizes the prediction back to the original resolution. This gives the best quality, as Stable Diffusion, from which Marigold is derived, performs best at 768x768 resolution.  
+- By default, the inference script resizes input images to the *processing resolution*, and then resizes the prediction back to the original resolution. This gives the best quality, as Stable Diffusion, from which SCFDepth is derived, performs best at 768x768 resolution.  
   
   - `--processing_res`: the processing resolution; set as 0 to process the input resolution directly. When unassigned (`None`), will read default setting from model config. Default: ~~768~~ `None`.
   - `--output_processing_res`: produce output at the processing resolution instead of upsampling it to the input resolution. Default: False.
@@ -111,28 +120,25 @@ The default settings are optimized for the best result. However, the behavior of
 
 ### ⬇ Checkpoint cache
 
-By default, the [checkpoint](https://huggingface.co/prs-eth/marigold-v1-0) is stored in the Hugging Face cache.
+By default, the [checkpoint](https://huggingface.co/developy/ApDepth) is stored in the Hugging Face cache.
 The `HF_HOME` environment variable defines its location and can be overridden, e.g.:
 
 ```bash
 export HF_HOME=$(pwd)/cache
 ```
-
+<!-- 
 Alternatively, use the following script to download the checkpoint weights locally:
 
 ```bash
 bash script/download_weights.sh marigold-v1-0
-# or LCM checkpoint
-bash script/download_weights.sh marigold-lcm-v1-0
-```
+``` -->
 
 At inference, specify the checkpoint path:
 
 ```bash
 python run.py \
-    --checkpoint checkpoint/marigold-v1-0 \
-    --denoise_steps 50 \
-    --ensemble_size 10 \
+    --checkpoint checkpoints/SCFDepth \
+    --ensemble_size 1 \
     --input_rgb_dir input/in-the-wild_example\
     --output_dir output/in-the-wild_example
 ```
@@ -163,44 +169,27 @@ bash script/eval/11_infer_nyu.sh
 bash script/eval/12_eval_nyu.sh
 ```
 
-Note: although the seed has been set, the results might still be slightly different on different hardware.
-
-## 🏋️ Training
-
-Based on the previously created environment, install extended requirements:
-
-```bash
-pip install -r requirements++.txt -r requirements+.txt -r requirements.txt
-```
-
-Set environment parameters for the data directory:
+Alternatively, use the following script to evaluate all datasets.
 
 ```bash
-export BASE_DATA_DIR=YOUR_DATA_DIR  # directory of training data
-export BASE_CKPT_DIR=YOUR_CHECKPOINT_DIR  # directory of pretrained checkpoint
+bash script/eval/00_test_all.sh
 ```
+You can get the result under `output/eval`
 
-Download Stable Diffusion v2 [checkpoint](https://huggingface.co/stabilityai/stable-diffusion-2) into `${BASE_CKPT_DIR}`
+> [!IMPORTANT]
+>
+> Although the seed has been set, the results might still be slightly different on different hardware.
 
-Prepare for [Hypersim](https://github.com/apple/ml-hypersim) and [Virtual KITTI 2](https://europe.naverlabs.com/research/computer-vision/proxy-virtual-worlds-vkitti-2/) datasets and save into `${BASE_DATA_DIR}`. Please refer to [this README](script/dataset_preprocess/hypersim/README.md) for Hypersim preprocessing.
 
-Run training script
+**Evaluating results**
 
-```bash
-python train.py --config config/train_marigold.yaml --no_wandb
-```
+Only the U-Net is updated and saved during training. To use the inference pipeline with your training result, replace `unet` folder in `train_SCFDepth` checkpoints with that in the `checkpoint` output folder. Then refer to [this section](#evaluation) for evaluation.
 
-Resume from a checkpoint, e.g.
+> [!IMPORTANT]
+>
+> Although random seeds have been set, the training result might be slightly different on different hardwares. It's recommended to train without interruption.
 
-```bash
-python train.py --resume_run output/train_marigold/checkpoint/latest --no_wandb
-```
-
-Evaluating results
-
-Only the U-Net is updated and saved during training. To use the inference pipeline with your training result, replace `unet` folder in Marigold checkpoints with that in the `checkpoint` output folder. Then refer to [this section](#evaluation) for evaluation.
 
-**Note**: Although random seeds have been set, the training result might be slightly different on different hardwares. It's recommended to train without interruption.
 
 ## ✏️ Contributing
 
@@ -215,17 +204,16 @@ Please refer to [this](CONTRIBUTING.md) instruction.
 
 
 ## 🎓 Citation
-Waitting for publishing⏱️
-<!-- Please cite our paper:
+Please cite our paper:
 
 ```bibtex
-@InProceedings{ke2023repurposing,
-      title={Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation},
-      author={Bingxin Ke and Anton Obukhov and Shengyu Huang and Nando Metzger and Rodrigo Caye Daudt and Konrad Schindler},
-      booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
-      year={2024}
+@InProceedings{haruko26scfdepth,
+      title={SCFDepth: A Single-Step Coarse-to-Fine Diffusion Framework for Monocular Depth Estimation},
+      author={Haruko386 and Yuan Shuai},
+      booktitle = {Under review},
+      year={2026}
 }
-``` -->
+```
 
 ## 🎫 License