ready to release

.gitignore

@@ -153,10 +153,8 @@ fabric.properties
 data
 public_data
 exp
-tmp
-exp_old
+eval
 
 script.sh
 get_exp.sh
-eval
 tmp.sh

README.md

@@ -7,7 +7,7 @@ This is the official implementation of **RNb-NeuS: Reflectance and Normal-based
 [Jean Mélou](https://www.irit.fr/~Jean.Melou/),
 [François Lauze](https://loutchoa.github.io/),
 [Jean-Denis Durou](https://www.irit.fr/~Jean-Denis.Durou/),
-[Lilian Calvet](https://www.youtube.com/watch?v=dQw4w9WgXcQ&pp=ygUKcmljayBhc2xleQ%3D%3D)
+[Lilian Calvet](https://scholar.google.com/citations?user=6JewdrMAAAAJ&hl=en)
 
 ### [Project page](https://robinbruneau.github.io/publications/rnb_neus.html) | [Paper](https://arxiv.org/abs/2312.01215)
 
@@ -24,7 +24,52 @@ pip install -r requirements.txt
 
 ## Usage
 
-Coming later...
+#### Data Convention
+
+Our data format is inspired from [IDR](https://github.com/lioryariv/idr/blob/main/DATA_CONVENTION.md) as follows:
+```
+CASE_NAME
+|-- cameras_xxx.npz    # camera parameters
+|-- normal
+    |-- 000.png        # normal map for each view
+    |-- 001.png
+    ...
+|-- albedo
+    |-- 000.png        # albedo for each view (optional)
+    |-- 001.png
+    ...
+|-- mask
+    |-- 000.png        # mask for each view
+    |-- 001.png
+    ...
+```
+
+One can create folders with different data in it, for instance, a normal folder for each normal estimation method.
+The name of the folder must be set in the used `.conf` file.
+
+We provide the [DiLiGenT-MV](https://drive.google.com/file/d/1TEBM6Dd7IwjRqJX0p8JwT9hLmy_vA5nU/view?usp=drive_link) data as described above with normals and reflectance maps estimated with [SDM-UniPS](https://github.com/satoshi-ikehata/SDM-UniPS-CVPR2023/). Note that the reflectance maps were scaled over all views and uncertainty masks were generated from 100 normals estimations (see the article for further details).
+
+### Run RNb-NeuS!
+
+**Train with reflectance**
+
+```shell
+python exp_runner.py --mode train_rnb --conf ./confs/CONF_NAME.conf --case CASE_NAME
+```
+
+**Train without reflectance**
+
+```shell
+python exp_runner.py --mode train_rnb --conf ./confs/CONF_NAME.conf --case CASE_NAME --no_albedo
+```
+
+**Extract surface** 
+
+```shell
+python exp_runner.py --mode validate_mesh --conf ./confs/CONF_NAME.conf --case CASE_NAME --is_continue
+```
+
+Additionaly, we provide the five meshes of the DiLiGenT-MV dataset with our method [here](https://drive.google.com/file/d/1CTQW1YLWOT2sSEWznFmSY_cUUtiTXLdM/view?usp=drive_link).
 
 ## Citation
 If you find our code useful for your research, please cite
@@ -36,4 +81,4 @@ If you find our code useful for your research, please cite
     archivePrefix={arXiv},
     year={2023}
 }
-```
+```

confs/wmask_rnb.conf

@@ -11,7 +11,7 @@ dataset {
 
     normal_dir = normal
     albedo_dir = albedo
-    mask_dir = mask_geodist
+    mask_dir = mask
 
     render_cameras_name = cameras.npz
     object_cameras_name = cameras.npz

confs/wmask_rnb_noalbedo.conf

@@ -10,7 +10,7 @@ dataset {
     data_dir = ./data/CASE_NAME/
 
     normal_dir = normal
-    mask_dir = mask_geodist
+    mask_dir = mask
 
     render_cameras_name = cameras.npz
     object_cameras_name = cameras.npz

confs/womask_rnb.conf

@@ -0,0 +1,89 @@
+general {
+    base_exp_dir = ./exp/CASE_NAME/womask_rnb
+    recording = [
+        ./,
+        ./models
+    ]
+}
+
+dataset {
+    data_dir = ./data/CASE_NAME/
+
+    normal_dir = normal
+    albedo_dir = albedo
+
+    render_cameras_name = cameras.npz
+    object_cameras_name = cameras.npz
+}
+
+train {
+    learning_rate = 5e-4,
+    learning_rate_alpha = 0.05,
+    end_iter = 300000,
+    warm_up_iter = 200000,
+
+    batch_size = 512,
+    validate_resolution_level = 4,
+    warm_up_end = 5000,
+    anneal_end = 50000,
+    use_white_bkgd = False,
+
+    save_freq = 10000,
+    val_freq = 5000,
+    val_mesh_freq = 25000,
+    report_freq = 500,
+
+    igr_weight = 0.1,
+    mask_weight = 0,
+}
+
+model {
+    nerf {
+        D = 8,
+        d_in = 4,
+        d_in_view = 3,
+        W = 256,
+        multires = 10,
+        multires_view = 4,
+        output_ch = 4,
+        skips=[4],
+        use_viewdirs=True
+    }
+
+    sdf_network {
+        d_out = 257,
+        d_in = 3,
+        d_hidden = 256,
+        n_layers = 8,
+        skip_in = [4],
+        multires = 6,
+        bias = 0.5,
+        scale = 1.0,
+        geometric_init = True,
+        weight_norm = True
+    }
+
+    variance_network {
+        init_val = 0.3
+    }
+
+    rendering_network {
+        d_feature = 256,
+        mode = no_view_dir,
+        d_in = 6,
+        d_out = 3,
+        d_hidden = 256,
+        n_layers = 2,
+        weight_norm = True,
+        multires_view = 4,
+        squeeze_out = True
+    }
+
+    neus_renderer {
+        n_samples = 64,
+        n_importance = 64,
+        n_outside = 32,
+        up_sample_steps = 4,    # 1 for simple coarse-to-fine sampling
+        perturb = 1.0
+    }
+}

confs/womask_rnb_noalbedo.conf

@@ -0,0 +1,88 @@
+general {
+    base_exp_dir = ./exp/CASE_NAME/womask_rnb_noalbedo
+    recording = [
+        ./,
+        ./models
+    ]
+}
+
+dataset {
+    data_dir = ./data/CASE_NAME/
+
+    normal_dir = normal
+
+    render_cameras_name = cameras.npz
+    object_cameras_name = cameras.npz
+}
+
+train {
+    learning_rate = 5e-4,
+    learning_rate_alpha = 0.05,
+    end_iter = 300000,
+    warm_up_iter = 200000,
+
+    batch_size = 512,
+    validate_resolution_level = 4,
+    warm_up_end = 5000,
+    anneal_end = 50000,
+    use_white_bkgd = False,
+
+    save_freq = 10000,
+    val_freq = 5000,
+    val_mesh_freq = 25000,
+    report_freq = 500,
+
+    igr_weight = 0.1,
+    mask_weight = 0,
+}
+
+model {
+    nerf {
+        D = 8,
+        d_in = 4,
+        d_in_view = 3,
+        W = 256,
+        multires = 10,
+        multires_view = 4,
+        output_ch = 4,
+        skips=[4],
+        use_viewdirs=True
+    }
+
+    sdf_network {
+        d_out = 257,
+        d_in = 3,
+        d_hidden = 256,
+        n_layers = 8,
+        skip_in = [4],
+        multires = 6,
+        bias = 0.5,
+        scale = 1.0,
+        geometric_init = True,
+        weight_norm = True
+    }
+
+    variance_network {
+        init_val = 0.3
+    }
+
+    rendering_network {
+        d_feature = 256,
+        mode = no_view_dir,
+        d_in = 6,
+        d_out = 3,
+        d_hidden = 256,
+        n_layers = 2,
+        weight_norm = True,
+        multires_view = 4,
+        squeeze_out = True
+    }
+
+    neus_renderer {
+        n_samples = 64,
+        n_importance = 64,
+        n_outside = 32,
+        up_sample_steps = 4,    # 1 for simple coarse-to-fine sampling
+        perturb = 1.0
+    }
+}