wenmin-wu

wenmin-wu/ds-skills

Skill Claude CodeCodex

Converts a pretrained 2D CNN into a 3D CNN by recursively replacing Conv2d/BN2d/Pool2d layers and inflating kernel weights along the depth axis.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Select center slices from a 3D volume and zero-pad along depth when the scan has fewer slices than the target count.

wenmin-wu/ds-skills

Skill Claude CodeCodex

3D ResNet encoder extracts volumetric features, pools depth dimension, then feeds into a 2D UNet/FPN decoder for segmentation.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Applies Mixup augmentation to 3D volumetric images and their segmentation masks, interpolating both inputs and loss targets.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Tiles 3D volumes into overlapping patches for inference and averages overlapping regions to produce seamless predictions.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Applies albumentations augmentations to object detection data while preserving bbox-label correspondence via BboxParams.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Resample a 3D medical volume to a fixed network input shape using physical voxel spacing (dz, dy, dx), correcting the Z dimension by the dz/dy ratio so anisotropic CT scans (1mm in-plane, 5mm slice) end up anatomically isotropic before trilinear interpolation.

wenmin-wu/ds-skills

Skill Claude CodeCodex

ArcFace angular margin loss layer for learning discriminative embeddings — used in image retrieval, product matching, and face recognition.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Show-Attend-Tell decoder: Bahdanau additive attention over CNN spatial features driving a gated LSTMCell for autoregressive image-to-sequence generation.

wenmin-wu/ds-skills

Skill Claude CodeCodex

All-vs-all contrastive loss comparing every pair in a batch (N^2 pairs) with margin and compactification regularizer.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Wrap albumentations bbox transforms in a 10-attempt retry loop that re-rolls the augmentation when all boxes get cropped out, preventing empty-target samples from poisoning the detection loss with NaN gradients.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Interpolate missing bounding boxes across video frames using bidirectional pandas interpolation to maintain smooth tracking through occlusions.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Combines BCE-with-logits and soft Dice loss with configurable weights for binary and multilabel segmentation training.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Replace a pretrained ResNet's conv1 with a wider input conv to accept stacked BEV raster channels (semantic map + agent history) while keeping downstream weights.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Two-stage CT classifier where a 2D CNN dumps per-slice features once, then a bidirectional GRU runs over the slice sequence to produce both per-slice predictions (TimeDistributed head) and an exam-level prediction (avg+max pooled head) — turns expensive 3D CNN training into cheap sequence modeling.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Crops uninformative black or dark borders from images by deriving a binary mask from the grayscale channel and trimming rows/columns.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Detect all-black DICOM/MRI slices (mean==0) and replace them by randomly sampling a non-black slice from the same series.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Fuse low-resolution multispectral bands with a high-resolution panchromatic band using the Brovey transform to produce sharp multi-band imagery.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Select a fixed number of Z-slices centered around the volume midpoint for memory-efficient 2.5D input from 3D CT/MRI stacks.

wenmin-wu/ds-skills

Skill Claude CodeCodex

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.

wenmin-wu/ds-skills

Skill Claude CodeCodex

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.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Compute pairwise cosine similarity on GPU in fixed-size chunks to avoid OOM, transferring only threshold-passing results to CPU.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Dynamically undersample majority class each epoch with per-class keep probabilities for stochastic balance.

wenmin-wu/ds-skills

Skill Claude CodeCodex

Sample training patches from large images using per-class area-fraction thresholds to ensure each patch contains meaningful object coverage.