メニュー
Stack 1D convolutions for local feature extraction before bidirectional LSTMs to classify variable-length stroke sequences into hundreds of doodle categories
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | cv-conv1d-lstm-stroke-classifier |
| description | Stack 1D convolutions for local feature extraction before bidirectional LSTMs to classify variable-length stroke sequences into hundreds of doodle categories |
Sketches can be represented as sequences of (x, y, stroke_id) points rather than rasterized images. A Conv1D-LSTM architecture processes this raw sequence: 1D convolutions extract local patterns (corners, curves), then LSTMs capture long-range stroke order dependencies. This approach works directly on stroke coordinates without rendering, preserving resolution-independent information.
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import (Conv1D, LSTM, Dense,
Dropout, BatchNormalization)
n_classes = 340
max_points = 128
model = Sequential([
BatchNormalization(input_shape=(max_points, 3)),
Conv1D(48, 5, activation="relu"),
Dropout(0.3),
Conv1D(64, 5, activation="relu"),
Dropout(0.3),
Conv1D(96, 3, activation="relu"),
Dropout(0.3),
LSTM(128, return_sequences=True),
Dropout(0.3),
LSTM(128, return_sequences=False),
Dropout(0.3),
Dense(512, activation="relu"),
Dropout(0.3),
Dense(n_classes, activation="softmax"),
])
model.compile(optimizer="adam", loss="categorical_crossentropy",
metrics=["accuracy"])
Memory-efficient Keras generator that streams sharded CSV files in chunks, renders strokes to images on-the-fly, and yields batches for training on datasets too large for memory
Partition a large dataset into N balanced shards using integer key modulo arithmetic for reproducible, class-interleaved splits across CSV files
Normalize raw stroke coordinates to 0-255 range, resample at uniform arc-length spacing, then apply Ramer-Douglas-Peucker simplification
Render stroke sequences to grayscale images with temporal intensity encoding where earlier strokes are brighter and later strokes fade to encode drawing order
Score each face in a video as anomalous by computing L2 distance from the embedding centroid of all faces, then convert to probability via logistic function
Extract per-frame CNN features then classify the temporal sequence with stacked GRU layers and a boolean mask for variable-length video inputs