Fix backward compatability of pos_enc bug fix

configs/loftr/indoor/buggy_pos_enc/loftr_ds.py

@@ -0,0 +1,6 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
+
+cfg.TRAINER.MSLR_MILESTONES = [3, 6, 9, 12, 17, 20, 23, 26, 29]

configs/loftr/indoor/buggy_pos_enc/loftr_ds_dense.py

@@ -0,0 +1,8 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
+
+cfg.LOFTR.MATCH_COARSE.SPARSE_SPVS = False
+
+cfg.TRAINER.MSLR_MILESTONES = [3, 6, 9, 12, 17, 20, 23, 26, 29]

configs/loftr/indoor/buggy_pos_enc/loftr_ot.py

@@ -0,0 +1,6 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'sinkhorn'
+
+cfg.TRAINER.MSLR_MILESTONES = [3, 6, 9, 12, 17, 20, 23, 26, 29]

configs/loftr/indoor/buggy_pos_enc/loftr_ot_dense.py

@@ -0,0 +1,8 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'sinkhorn'
+
+cfg.LOFTR.MATCH_COARSE.SPARSE_SPVS = False
+
+cfg.TRAINER.MSLR_MILESTONES = [3, 6, 9, 12, 17, 20, 23, 26, 29]

configs/loftr/indoor/scannet/loftr_ds_eval.py

@@ -10,6 +10,7 @@ to be consistent with the results in our paper.
 
 from src.config.default import _CN as cfg
 
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
 cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
 
 cfg.LOFTR.MATCH_COARSE.BORDER_RM = 0

configs/loftr/indoor/scannet/loftr_ds_eval_new.py

@@ -0,0 +1,18 @@
+""" A config only for reproducing the ScanNet evaluation results.
+
+We remove border matches by default, but the originally implemented
+`remove_border()` has a bug, leading to only two sides of
+all borders are actually removed. However, the [bug fix](https://github.com/zju3dv/LoFTR/commit/e9146c8144dea5f3cbdd98b225f3e147a171c216)
+makes the scannet evaluation results worse (auc@10=40.8 => 39.5), which should be
+caused by tiny result fluctuation of few image pairs. This config set `BORDER_RM` to 0
+to be consistent with the results in our paper.
+
+Update: This config is for testing the re-trained model with the pos-enc bug fixed.
+"""
+
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = True
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
+
+cfg.LOFTR.MATCH_COARSE.BORDER_RM = 0

configs/loftr/outdoor/buggy_pos_enc/loftr_ds.py

@@ -0,0 +1,16 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
+
+cfg.TRAINER.CANONICAL_LR = 8e-3
+cfg.TRAINER.WARMUP_STEP = 1875  # 3 epochs
+cfg.TRAINER.WARMUP_RATIO = 0.1
+cfg.TRAINER.MSLR_MILESTONES = [8, 12, 16, 20, 24]
+
+# pose estimation
+cfg.TRAINER.RANSAC_PIXEL_THR = 0.5
+
+cfg.TRAINER.OPTIMIZER = "adamw"
+cfg.TRAINER.ADAMW_DECAY = 0.1
+cfg.LOFTR.MATCH_COARSE.TRAIN_COARSE_PERCENT = 0.3

configs/loftr/outdoor/buggy_pos_enc/loftr_ds_dense.py

@@ -0,0 +1,17 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'dual_softmax'
+cfg.LOFTR.MATCH_COARSE.SPARSE_SPVS = False
+
+cfg.TRAINER.CANONICAL_LR = 8e-3
+cfg.TRAINER.WARMUP_STEP = 1875  # 3 epochs
+cfg.TRAINER.WARMUP_RATIO = 0.1
+cfg.TRAINER.MSLR_MILESTONES = [8, 12, 16, 20, 24]
+
+# pose estimation
+cfg.TRAINER.RANSAC_PIXEL_THR = 0.5
+
+cfg.TRAINER.OPTIMIZER = "adamw"
+cfg.TRAINER.ADAMW_DECAY = 0.1
+cfg.LOFTR.MATCH_COARSE.TRAIN_COARSE_PERCENT = 0.3

configs/loftr/outdoor/buggy_pos_enc/loftr_ot.py

@@ -0,0 +1,16 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'sinkhorn'
+
+cfg.TRAINER.CANONICAL_LR = 8e-3
+cfg.TRAINER.WARMUP_STEP = 1875  # 3 epochs
+cfg.TRAINER.WARMUP_RATIO = 0.1
+cfg.TRAINER.MSLR_MILESTONES = [8, 12, 16, 20, 24]
+
+# pose estimation
+cfg.TRAINER.RANSAC_PIXEL_THR = 0.5
+
+cfg.TRAINER.OPTIMIZER = "adamw"
+cfg.TRAINER.ADAMW_DECAY = 0.1
+cfg.LOFTR.MATCH_COARSE.TRAIN_COARSE_PERCENT = 0.3

configs/loftr/outdoor/buggy_pos_enc/loftr_ot_dense.py

@@ -0,0 +1,17 @@
+from src.config.default import _CN as cfg
+
+cfg.LOFTR.COARSE.TEMP_BUG_FIX = False
+cfg.LOFTR.MATCH_COARSE.MATCH_TYPE = 'sinkhorn'
+cfg.LOFTR.MATCH_COARSE.SPARSE_SPVS = False
+
+cfg.TRAINER.CANONICAL_LR = 8e-3
+cfg.TRAINER.WARMUP_STEP = 1875  # 3 epochs
+cfg.TRAINER.WARMUP_RATIO = 0.1
+cfg.TRAINER.MSLR_MILESTONES = [8, 12, 16, 20, 24]
+
+# pose estimation
+cfg.TRAINER.RANSAC_PIXEL_THR = 0.5
+
+cfg.TRAINER.OPTIMIZER = "adamw"
+cfg.TRAINER.ADAMW_DECAY = 0.1
+cfg.LOFTR.MATCH_COARSE.TRAIN_COARSE_PERCENT = 0.3

scripts/reproduce_test/indoor_ds_new.sh

@@ -0,0 +1,30 @@
+#!/bin/bash -l
+# a indoor_ds model with the pos_enc impl bug fixed.
+
+SCRIPTPATH=$(dirname $(readlink -f "$0"))
+PROJECT_DIR="${SCRIPTPATH}/../../"
+
+# conda activate loftr
+export PYTHONPATH=$PROJECT_DIR:$PYTHONPATH
+cd $PROJECT_DIR
+
+data_cfg_path="configs/data/scannet_test_1500.py"
+main_cfg_path="configs/loftr/indoor/scannet/loftr_ds_eval_new.py"
+ckpt_path="weights/indoor_ds_new.ckpt"
+dump_dir="dump/loftr_ds_indoor_new"
+profiler_name="inference"
+n_nodes=1  # mannually keep this the same with --nodes
+n_gpus_per_node=-1
+torch_num_workers=4
+batch_size=1  # per gpu
+
+python -u ./test.py \
+    ${data_cfg_path} \
+    ${main_cfg_path} \
+    --ckpt_path=${ckpt_path} \
+    --dump_dir=${dump_dir} \
+    --gpus=${n_gpus_per_node} --num_nodes=${n_nodes} --accelerator="ddp" \
+    --batch_size=${batch_size} --num_workers=${torch_num_workers}\
+    --profiler_name=${profiler_name} \
+    --benchmark 
+    

scripts/reproduce_test/indoor_ot.sh

@@ -8,7 +8,7 @@ export PYTHONPATH=$PROJECT_DIR:$PYTHONPATH
 cd $PROJECT_DIR
 
 data_cfg_path="configs/data/scannet_test_1500.py"
-main_cfg_path="configs/loftr/indoor/loftr_ot.py"
+main_cfg_path="configs/loftr/indoor/buggy_pos_enc/loftr_ot.py"
 ckpt_path="weights/indoor_ot.ckpt"
 dump_dir="dump/loftr_ot_indoor"
 profiler_name="inference"

scripts/reproduce_test/outdoor_ds.sh

@@ -8,7 +8,7 @@ export PYTHONPATH=$PROJECT_DIR:$PYTHONPATH
 cd $PROJECT_DIR
 
 data_cfg_path="configs/data/megadepth_test_1500.py"
-main_cfg_path="configs/loftr/outdoor/loftr_ds.py"
+main_cfg_path="configs/loftr/outdoor/buggy_pos_enc/loftr_ds.py"
 ckpt_path="weights/outdoor_ds.ckpt"
 dump_dir="dump/loftr_ds_outdoor"
 profiler_name="inference"

scripts/reproduce_test/outdoor_ot.sh

@@ -8,7 +8,7 @@ export PYTHONPATH=$PROJECT_DIR:$PYTHONPATH
 cd $PROJECT_DIR
 
 data_cfg_path="configs/data/megadepth_test_1500.py"
-main_cfg_path="configs/loftr/outdoor/loftr_ot.py"
+main_cfg_path="configs/loftr/outdoor/buggy_pos_enc/loftr_ot.py"
 ckpt_path="weights/outdoor_ot.ckpt"
 dump_dir="dump/loftr_ot_outdoor"
 profiler_name="inference"

src/config/default.py

@@ -20,6 +20,7 @@ _CN.LOFTR.COARSE.D_FFN = 256
 _CN.LOFTR.COARSE.NHEAD = 8
 _CN.LOFTR.COARSE.LAYER_NAMES = ['self', 'cross'] * 4
 _CN.LOFTR.COARSE.ATTENTION = 'linear'  # options: ['linear', 'full']
+_CN.LOFTR.COARSE.TEMP_BUG_FIX = True
 
 # 3. Coarse-Matching config
 _CN.LOFTR.MATCH_COARSE = CN()

src/lightning/lightning_loftr.py

@@ -44,7 +44,11 @@ class PL_LoFTR(pl.LightningModule):
 
         # Pretrained weights
         if pretrained_ckpt:
-            self.matcher.load_state_dict(torch.load(pretrained_ckpt, map_location='cpu')['state_dict'])
+            state_dict = torch.load(pretrained_ckpt, map_location='cpu')['state_dict']
+            for k in list(state_dict.keys()):
+                if k.startswith('matcher.'):
+                    state_dict[k.replace('matcher.', '', 1)] = state_dict.pop(k)
+            self.matcher.load_state_dict(state_dict, strict=True)
             logger.info(f"Load \'{pretrained_ckpt}\' as pretrained checkpoint")
         
         # Testing

src/loftr/loftr.py

@@ -17,7 +17,9 @@ class LoFTR(nn.Module):
 
         # Modules
         self.backbone = build_backbone(config)
-        self.pos_encoding = PositionEncodingSine(config['coarse']['d_model'])
+        self.pos_encoding = PositionEncodingSine(
+            config['coarse']['d_model'],
+            temp_bug_fix=config['coarse']['temp_bug_fix'])
         self.loftr_coarse = LocalFeatureTransformer(config['coarse'])
         self.coarse_matching = CoarseMatching(config['match_coarse'])
         self.fine_preprocess = FinePreprocess(config)

src/loftr/utils/position_encoding.py

@@ -8,17 +8,24 @@ class PositionEncodingSine(nn.Module):
     This is a sinusoidal position encoding that generalized to 2-dimensional images
     """
 
-    def __init__(self, d_model, max_shape=(256, 256)):
+    def __init__(self, d_model, max_shape=(256, 256), temp_bug_fix=True):
         """
         Args:
             max_shape (tuple): for 1/8 featmap, the max length of 256 corresponds to 2048 pixels
+            temp_bug_fix (bool): As noted in this [issue](https://github.com/zju3dv/LoFTR/issues/41),
+                the original implementation of LoFTR includes a bug in the pos-enc impl, which has little impact
+                on the final performance. For now, we keep both impls for backward compatability.
+                We will remove the buggy impl after re-training all variants of our released models.
         """
         super().__init__()
 
         pe = torch.zeros((d_model, *max_shape))
         y_position = torch.ones(max_shape).cumsum(0).float().unsqueeze(0)
         x_position = torch.ones(max_shape).cumsum(1).float().unsqueeze(0)
-        div_term = torch.exp(torch.arange(0, d_model//2, 2).float() * (-math.log(10000.0) / (d_model//2)))
+        if temp_bug_fix:
+            div_term = torch.exp(torch.arange(0, d_model//2, 2).float() * (-math.log(10000.0) / (d_model//2)))
+        else:  # a buggy implementation (for backward compatability only)
+            div_term = torch.exp(torch.arange(0, d_model//2, 2).float() * (-math.log(10000.0) / d_model//2))
         div_term = div_term[:, None, None]  # [C//4, 1, 1]
         pe[0::4, :, :] = torch.sin(x_position * div_term)
         pe[1::4, :, :] = torch.cos(x_position * div_term)