synthia 3d weights

chrischoy · Jan 30, 2020 · f526c2a · f526c2a
1 parent 1c937e8
commit f526c2a
Show file tree

Hide file tree

Showing 3 changed files with 20 additions and 6 deletions.
diff --git a/README.md b/README.md
@@ -133,12 +133,14 @@ python -m lib.datasets.preprocessing.stanford
 
 ## Model Zoo
 
-| Model         | Dataset             | Voxel Size | Conv1 Kernel Size | Performance              | Link   |
-|:-------------:|:-------------------:|:----------:|:-----------------:|:-------------------------------------------------------------------------------:|:------:|
-| Mink16UNet34C | ScanNet train + val | 2cm        | 3                 | Test set 73.6% mIoU, no sliding window                                          | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet34C_ScanNet.pth) |
-| Mink16UNet34C | ScanNet train       | 2cm        | 5                 | Val 72.219% mIoU, no rotation average, no sliding window [per class performance](https://github.com/chrischoy/SpatioTemporalSegmentation/issues/13) | [download](https://node1.chrischoy.org/data/publications/minknet/MinkUNet34C-train-conv1-5.pth) |
-| Mink16UNet18  | Stanford Area5 train | 5cm       | 5                 | Area 5 test 65.828% mIoU, no rotation average, no sliding window [per class performance](https://pastebin.com/Gj3PrPFr) | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet18-stanford-conv1-5.pth) |
-| Mink16UNet34  | Stanford Area5 train | 5cm       | 5                 | Area 5 test 66.348% mIoU, no rotation average, no sliding window [per class performance](https://pastebin.com/WzhfGMQG) | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet34-stanford-conv1-5.pth) |
+| Model            | Dataset             | Voxel Size | Conv1 Kernel Size | Performance              | Link   |
+|:----------------:|:-------------------:|:----------:|:-----------------:|:-------------------------------------------------------------------------------:|:------:|
+| Mink16UNet34C    | ScanNet train + val | 2cm        | 3                 | Test set 73.6% mIoU, no sliding window                                          | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet34C_ScanNet.pth) |
+| Mink16UNet34C    | ScanNet train       | 2cm        | 5                 | Val 72.219% mIoU, no rotation average, no sliding window [per class performance](https://github.com/chrischoy/SpatioTemporalSegmentation/issues/13) | [download](https://node1.chrischoy.org/data/publications/minknet/MinkUNet34C-train-conv1-5.pth) |
+| Mink16UNet18     | Stanford Area5 train | 5cm       | 5                 | Area 5 test 65.828% mIoU, no rotation average, no sliding window [per class performance](https://pastebin.com/Gj3PrPFr) | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet18-stanford-conv1-5.pth) |
+| Mink16UNet34     | Stanford Area5 train | 5cm       | 5                 | Area 5 test 66.348% mIoU, no rotation average, no sliding window [per class performance](https://pastebin.com/WzhfGMQG) | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet34-stanford-conv1-5.pth) |
+| 3D Mink16UNet14A | Synthia CVPR19 train | 15cm      | 3                 | CVPR19 test wo elastic distortion 81.903% mIoU, no rotation average, no sliding window [per class performance](https://pastebin.com/hN88iQqa) | [download](https://node1.chrischoy.org/data/publications/minknet/Mink16UNet14A-synthia-conv1-3.pth) |
+
 
 Note that sliding window style evaluation (cropping and stitching results) used in many related works effectively works as an ensemble (rotation averaging) which boosts the performance.
 

diff --git a/lib/datasets/synthia.py b/lib/datasets/synthia.py
@@ -265,6 +265,10 @@ def _transform(xyz, intrinsic, extrinsic):
     return ptc, center
 
 
+class SynthiaCVPR10cmVoxelizationDataset(SynthiaVoxelizationDataset):
+  VOXEL_SIZE = 10
+
+
 class SynthiaCVPR15cmVoxelizationDataset(SynthiaVoxelizationDataset):
   pass
 

diff --git a/scripts/train_synthia4d.sh b/scripts/train_synthia4d.sh
@@ -33,3 +33,11 @@ python -m main \
     --train_limit_numpoints 1200000 \
     --train_phase train \
     $3 2>&1 | tee -a "$LOG"
+
+python -m main \
+    --log_dir $LOG_DIR \
+    --dataset $DATASET \
+    --model $MODEL \
+    --is_train False \
+    --weights $LOG_DIR/weights.pth \
+    $3 2>&1 | tee -a "$LOG"