add training config

configs/instant-mesh-large-train.yaml

@@ -0,0 +1,68 @@
+model:
+  base_learning_rate: 4.0e-05
+  target: src.model_mesh.MVRecon
+  params:
+    init_ckpt: logs/instant-mesh-large/checkpoints/last.ckpt
+    input_size: 320
+    render_size: 512
+    
+    lrm_generator_config:
+      target: src.models.lrm_mesh.InstantMesh
+      params:
+        camera_embed_dim: 1024
+        rendering_samples_per_ray: 128
+        transformer_dim: 1024
+        transformer_layers: 16
+        transformer_heads: 16
+        triplane_low_res: 32
+        triplane_high_res: 64
+        triplane_dim: 80
+        encoder_feat_dim: 768
+        encoder_freeze: false
+        encoder_model_name: facebook/dino-vitb16
+        grid_res: 128
+        grid_scale: 2.1
+
+
+data:
+  target: src.data.objaverse.DataModuleFromConfig
+  params:
+    batch_size: 2
+    num_workers: 8
+    train:
+      target: src.data.objaverse.ObjaverseData
+      params:
+        root_dir: data/objaverse
+        meta_fname: filtered_obj_name.json
+        input_image_dir: rendering_random_32views
+        target_image_dir: rendering_random_32views
+        input_view_num: 6
+        target_view_num: 4
+        total_view_n: 32
+        fov: 50
+        camera_rotation: true
+        validation: false
+    validation:
+      target: src.data.objaverse.ValidationData
+      params:
+        root_dir: data/valid_samples
+        input_view_num: 6
+        input_image_size: 320
+        fov: 30
+
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 2000
+      save_top_k: -1
+      save_last: true
+  callbacks: {}
+
+  trainer:
+    benchmark: true
+    max_epochs: -1
+    val_check_interval: 1000
+    num_sanity_val_steps: 0
+    accumulate_grad_batches: 1
+    check_val_every_n_epoch: null   # if not set this, validation does not run

configs/instant-nerf-large-train.yaml

@@ -0,0 +1,66 @@
+model:
+  base_learning_rate: 4.0e-04
+  target: src.model.MVRecon
+  params:
+    input_size: 320
+    render_size: 192
+    
+    lrm_generator_config:
+      target: src.models.lrm.InstantNeRF
+      params:
+        camera_embed_dim: 1024
+        rendering_samples_per_ray: 128
+        transformer_dim: 1024
+        transformer_layers: 16
+        transformer_heads: 16
+        triplane_low_res: 32
+        triplane_high_res: 64
+        triplane_dim: 80
+        encoder_feat_dim: 768
+        encoder_freeze: false
+        encoder_model_name: facebook/dino-vitb16
+
+
+data:
+  target: src.data.objaverse.DataModuleFromConfig
+  params:
+    batch_size: 2
+    num_workers: 8
+    train:
+      target: src.data.objaverse.ObjaverseData
+      params:
+        root_dir: data/objaverse
+        meta_fname: filtered_obj_name.json
+        input_image_dir: rendering_random_32views
+        target_image_dir: rendering_random_32views
+        input_view_num: 6
+        target_view_num: 4
+        total_view_n: 32
+        fov: 50
+        camera_rotation: true
+        validation: false
+    validation:
+      target: src.data.objaverse.ValidationData
+      params:
+        root_dir: data/valid_samples
+        input_view_num: 6
+        input_image_size: 320
+        fov: 30
+
+
+lightning:
+  modelcheckpoint:
+    params:
+      every_n_train_steps: 1000
+      save_top_k: -1
+      save_last: true
+  callbacks: {}
+
+  trainer:
+    benchmark: true
+    max_epochs: -1
+    gradient_clip_val: 1.0
+    val_check_interval: 1000
+    num_sanity_val_steps: 0
+    accumulate_grad_batches: 1
+    check_val_every_n_epoch: null   # if not set this, validation does not run

src/data/objaverse.py

@@ -20,7 +20,7 @@ from src.utils.train_util import instantiate_from_config
 from src.utils.camera_util import (
     FOV_to_intrinsics, 
     center_looking_at_camera_pose, 
-    get_circular_camera_poses,
+    get_surrounding_views,
 )
 
 
@@ -76,7 +76,7 @@ class ObjaverseData(Dataset):
         input_image_dir='rendering_random_32views',
         target_image_dir='rendering_random_32views',
         input_view_num=6,
-        target_view_num=2,
+        target_view_num=4,
         total_view_n=32,
         fov=50,
         camera_rotation=True,
@@ -97,7 +97,7 @@ class ObjaverseData(Dataset):
         paths = filtered_dict['good_objs']
         self.paths = paths
         
-        self.depth_scale = 5.0
+        self.depth_scale = 4.0
             
         total_objects = len(self.paths)
         print('============= length of dataset %d =============' % len(self.paths))
@@ -120,7 +120,6 @@ class ObjaverseData(Dataset):
         return image, alpha
     
     def __getitem__(self, index):
-        # load data
         while True:
             input_image_path = os.path.join(self.root_dir, self.input_image_dir, self.paths[index])
             target_image_path = os.path.join(self.root_dir, self.target_image_dir, self.paths[index])
@@ -210,7 +209,7 @@ class ObjaverseData(Dataset):
 
         # random scaling
         if np.random.rand() < 0.5:
-            scale = np.random.uniform(0.8, 1.0)
+            scale = np.random.uniform(0.7, 1.1)
             c2ws[:, :3, 3] *= scale
             depths *= scale
 
@@ -223,7 +222,7 @@ class ObjaverseData(Dataset):
             'input_alphas': alphas[:self.input_view_num],           # (6, 1, H, W) 
             'input_depths': depths[:self.input_view_num],           # (6, 1, H, W)
             'input_normals': normals[:self.input_view_num],         # (6, 3, H, W)
-            'input_c2ws': c2ws[:self.input_view_num],         # (6, 4, 4)
+            'input_c2ws': c2ws_input[:self.input_view_num],         # (6, 4, 4)
             'input_Ks': Ks[:self.input_view_num],                   # (6, 3, 3)
 
             # lrm generator input and supervision
@@ -233,8 +232,6 @@ class ObjaverseData(Dataset):
             'target_normals': normals[self.input_view_num:],        # (V, 3, H, W)
             'target_c2ws': c2ws[self.input_view_num:],              # (V, 4, 4)
             'target_Ks': Ks[self.input_view_num:],                  # (V, 3, 3)
-
-            'depth_available': 1,
         }
         return data
 
@@ -244,7 +241,7 @@ class ValidationData(Dataset):
         root_dir='objaverse/',
         input_view_num=6,
         input_image_size=256,
-        fov=30,
+        fov=50,
     ):
         self.root_dir = Path(root_dir)
         self.input_view_num = input_view_num
@@ -254,9 +251,9 @@ class ValidationData(Dataset):
         self.paths = sorted(os.listdir(self.root_dir))
         print('============= length of dataset %d =============' % len(self.paths))
 
-        cam_distance = 4.0
+        cam_distance = 2.5
         azimuths = np.array([30, 90, 150, 210, 270, 330])
-        elevations = np.array([20, -10, 20, -10, 20, -10])
+        elevations = np.array([30, -20, 30, -20, 30, -20])
         azimuths = np.deg2rad(azimuths)
         elevations = np.deg2rad(elevations)
 
@@ -270,7 +267,7 @@ class ValidationData(Dataset):
         self.c2ws = c2ws.float()
         self.Ks = FOV_to_intrinsics(self.fov).unsqueeze(0).repeat(6, 1, 1).float()
 
-        render_c2ws = get_circular_camera_poses(M=8, radius=cam_distance, elevation=20.0)
+        render_c2ws = get_surrounding_views(M=8, radius=cam_distance)
         render_Ks = FOV_to_intrinsics(self.fov).unsqueeze(0).repeat(render_c2ws.shape[0], 1, 1)
         self.render_c2ws = render_c2ws.float()
         self.render_Ks = render_Ks.float()

src/model.py

@@ -295,16 +295,9 @@ class MVRecon(pl.LightningModule):
 
         params = []
 
-        lrm_params_fast, lrm_params_slow = [], []
-        for n, p in self.lrm_generator.named_parameters():
-            if 'adaLN_modulation' in n or 'camera_embedder' in n:
-                lrm_params_fast.append(p)
-            else:
-                lrm_params_slow.append(p)
-        params.append({"params": lrm_params_fast, "lr": lr, "weight_decay": 0.01 })
-        params.append({"params": lrm_params_slow, "lr": lr / 10.0, "weight_decay": 0.01 })
+        params.append({"params": self.lrm_generator.parameters(), "lr": lr, "weight_decay": 0.01 })
 
         optimizer = torch.optim.AdamW(params, lr=lr, betas=(0.90, 0.95))
-        scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(optimizer, 3000, eta_min=lr/4)
+        scheduler = torch.optim.lr_scheduler.CosineAnnealingWarmRestarts(optimizer, 3000, eta_min=lr/10)
 
         return {'optimizer': optimizer, 'lr_scheduler': scheduler}