add facornet dataset, model and training task (#71)

.pre-commit-config.yaml

@@ -1,20 +1,20 @@
 repos:
   - repo: https://github.com/mwouts/jupytext
-    rev: v1.15.2
+    rev: v1.16.1
     hooks:
     - id: jupytext
       args: [--sync, --pipe, black]
   - repo: https://github.com/psf/black
-    rev: 23.10.1
+    rev: 24.3.0
     hooks:
     - id: black
   - repo: https://github.com/pycqa/flake8
-    rev: 6.1.0
+    rev: 7.0.0
     hooks:
     - id: flake8
       additional_dependencies: [Flake8-pyproject]
   - repo: https://github.com/pycqa/isort
-    rev: 5.12.0
+    rev: 5.13.2
     hooks:
       - id: isort
         name: isort (python)

ours/datasets/zhang.py

@@ -1,51 +1,6 @@
-from pathlib import Path
+from fiw import FIW
 
-import cv2
-import torch
-from torch.utils.data import Dataset
-from utils import Sample
 
-
-class FIW(Dataset):
-    def __init__(self, root_dir, sample_path, transform=None):
-        self.root_dir = root_dir
-        self.sample_path = sample_path
-        self.transform = transform
-        self.bias = 0
-        self.sample_list = self.load_sample()
-        print(f"Loaded {len(self.sample_list)} samples from {sample_path}")
-
-    def load_sample(self):
-        sample_list = []
-        lines = Path(self.sample_path).read_text().strip().split("\n")
-        for line in lines:
-            if len(line) < 1:
-                continue
-            tmp = line.split(" ")
-            sample = Sample(tmp[0], tmp[1], tmp[2], tmp[-2], tmp[-1])
-            sample_list.append(sample)
-        return sample_list
-
-    def __len__(self):
-        return len(self.sample_list)
-
-    def read_image(self, path):
-        # TODO: add to utils.py
-        img = cv2.imread(f"{self.root_dir}/{path}")
-        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
-        img = cv2.resize(img, (112, 112))
-        return img
-
-    def set_bias(self, bias):
-        self.bias = bias
-
-    def __getitem__(self, item):
-        # id, f1, f2, kin_relation, is_kin
-        sample = self.sample_list[item + self.bias]
-        img1, img2 = self.read_image(sample.f1), self.read_image(sample.f2)
-        if self.transform is not None:
-            img1, img2 = self.transform(img1), self.transform(img2)
-        is_kin = torch.tensor(int(sample.is_kin))
-        kin_id = Sample.NAME2LABEL[sample.kin_relation] if is_kin else 0
-        labels = (kin_id, is_kin)
-        return img1, img2, labels
+class FIWSCL(FIW):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)

ours/guild.yml

@@ -107,3 +107,26 @@
           target-type: link
         - operation: mtcf
           select: exp
+
+- model: facornet
+  operations:
+    train:
+      description: Reproduction of Kinship Representation Learning with Face Componential Relation (2023)
+      main: tasks.facornet train
+      sourcecode:
+        - utils.py
+        - losses.py
+        - models/facornet.py
+        - datasets/fiw.py
+        - datasets/facornet.py
+        - datasets/utils.py
+        - tasks/facornet.py
+      flags-import: all
+      flags-dest: args
+      output-scalars: '(\key):\s+(\value)'
+      requires:
+        - file: weights
+          target-type: link
+        - file: ../datasets/
+          target-type: link
+          rename: data

ours/losses.py

@@ -0,0 +1,14 @@
+import torch
+
+
+def facornet_contrastive_loss(x1, x2, beta=0.08):
+    m = 0.0
+    x1x2 = torch.cat([x1, x2], dim=0)
+    x2x1 = torch.cat([x2, x1], dim=0)
+    beta = (beta**2).sum([1, 2]) / 500
+    beta = torch.cat([beta, beta]).reshape(-1)
+    cosine_mat = torch.cosine_similarity(torch.unsqueeze(x1x2, dim=1), torch.unsqueeze(x1x2, dim=0), dim=2) / (beta + m)
+    mask = 1.0 - torch.eye(2 * x1.size(0)).to(x1.device)
+    numerators = torch.exp(torch.cosine_similarity(x1x2, x2x1, dim=1) / (beta + m))
+    denominators = torch.sum(torch.exp(cosine_mat) * mask, dim=1)
+    return -torch.mean(torch.log(numerators / denominators), dim=0)