Add independently reproduced HRNet-R90JT checkpoint

README.md

@@ -0,0 +1,20 @@
+---
+library_name: pytorch
+tags:
+  - hrnet
+  - facial-landmarks
+  - infant
+---
+
+# MMInfant HRNet-R90JT
+
+This repository provides the HRNet-R90JT checkpoint used for facial landmark extraction in MMInfant.
+
+The checkpoint was trained by the MMInfant authors following the public training procedure from [InfAnFace](https://github.com/ostadabbas/Infant-Facial-Landmark-Detection-and-Tracking). It differs from the HRNet-R90JT checkpoint currently distributed by the upstream project.
+
+The model architecture is based on [HRNet Facial Landmark Detection](https://github.com/HRNet/HRNet-Facial-Landmark-Detection).
+
+## Files
+
+- `hrnet-r90jt.pth`: model weights
+- `hrnet-r90jt.yaml`: inference configuration

hrnet-r90jt.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69caf92728d961b319159cc04ea44497b66718ad7cf99c8224b3e3c7b8a1a8e5
+size 39387431

hrnet-r90jt.yaml

@@ -0,0 +1,59 @@
+# Inference configuration for the independently reproduced HRNet-R90JT checkpoint.
+# Network dimensions match the configuration used by the MMInfant landmark pipeline.
+# Training datasets, optimizer settings, and pretrained-initialization paths are omitted.
+GPUS: (0, )
+MODEL:
+  NAME: hrnet
+  NUM_JOINTS: 68
+  INIT_WEIGHTS: false
+  PRETRAINED: ''
+  SIGMA: 1.0
+  IMAGE_SIZE:
+  - 256
+  - 256
+  HEATMAP_SIZE:
+  - 64
+  - 64
+  EXTRA:
+    FINAL_CONV_KERNEL: 1
+    STAGE2:
+      NUM_MODULES: 1
+      NUM_BRANCHES: 2
+      BLOCK: BASIC
+      NUM_BLOCKS:
+      - 4
+      - 4
+      NUM_CHANNELS:
+      - 18
+      - 36
+      FUSE_METHOD: SUM
+    STAGE3:
+      NUM_MODULES: 4
+      NUM_BRANCHES: 3
+      BLOCK: BASIC
+      NUM_BLOCKS:
+      - 4
+      - 4
+      - 4
+      NUM_CHANNELS:
+      - 18
+      - 36
+      - 72
+      FUSE_METHOD: SUM
+    STAGE4:
+      NUM_MODULES: 3
+      NUM_BRANCHES: 4
+      BLOCK: BASIC
+      NUM_BLOCKS:
+      - 4
+      - 4
+      - 4
+      - 4
+      NUM_CHANNELS:
+      - 18
+      - 36
+      - 72
+      - 144
+      FUSE_METHOD: SUM
+TEST:
+  BATCH_SIZE_PER_GPU: 8