Comparison of different ImageNet-based CNN models for classifying diabetic retinopathy images 8aa7de25b342c5fbe2f85dcf68918eb5614b92c8 2026-04-22T13:44:03Z 2026-04-22T13:44:03Z Vineet Kumar git@vineetk.net commit 8aa7de25b342c5fbe2f85dcf68918eb5614b92c8 parent 9b8cf40297e4e0bee41e3a7697bb063c04ae7335 Author: Vineet Kumar <git@vineetk.net> Date: Wed, 22 Apr 2026 09:44:03 -0400 add original proposal presentation and move into assignment/ 9b8cf40297e4e0bee41e3a7697bb063c04ae7335 2026-04-22T03:59:25Z 2026-04-22T03:59:25Z Vineet Kumar git@vineetk.net commit 9b8cf40297e4e0bee41e3a7697bb063c04ae7335 parent a54bcf2255337052808ea2312cbd418c6a5f89a4 Author: Vineet Kumar <git@vineetk.net> Date: Tue, 21 Apr 2026 23:59:25 -0400 add results (except pth files) a54bcf2255337052808ea2312cbd418c6a5f89a4 2026-04-22T03:58:24Z 2026-04-22T03:58:24Z Vineet Kumar git@vineetk.net commit a54bcf2255337052808ea2312cbd418c6a5f89a4 parent 443033a57e93ae73785c063116f54ec26b903beb Author: Vineet Kumar <git@vineetk.net> Date: Tue, 21 Apr 2026 23:58:24 -0400 add final report 443033a57e93ae73785c063116f54ec26b903beb 2026-04-22T03:09:21Z 2026-04-22T03:09:21Z Vineet Kumar git@vineetk.net commit 443033a57e93ae73785c063116f54ec26b903beb parent 88375c7767dbea45e9a1158f0aa180043a59ffae Author: Vineet Kumar <git@vineetk.net> Date: Tue, 21 Apr 2026 23:09:21 -0400 add a README.md file 88375c7767dbea45e9a1158f0aa180043a59ffae 2026-04-22T02:54:46Z 2026-04-22T02:54:46Z Vineet Kumar git@vineetk.net commit 88375c7767dbea45e9a1158f0aa180043a59ffae parent 3d5c9e1658ea8c8b06d9acaeff847647b58fa63a Author: Vineet Kumar <git@vineetk.net> Date: Tue, 21 Apr 2026 22:54:46 -0400 main: increase batch size 3d5c9e1658ea8c8b06d9acaeff847647b58fa63a 2026-04-20T14:28:06Z 2026-04-20T14:28:06Z Vineet Kumar git@vineetk.net commit 3d5c9e1658ea8c8b06d9acaeff847647b58fa63a parent 9dd8b3007e79f0634f868d405f5e7c48917305fd Author: Vineet Kumar <git@vineetk.net> Date: Mon, 20 Apr 2026 10:28:06 -0400 increase resolution to 384px and soften oversampler weights - Default img_size changed from 224 to 384 (already plumbed through build_transforms/build_dataloaders, just needed passing from main.py) - WeightedRandomSampler now uses sqrt(1/count) instead of 1/count to reduce over-prediction of minority classes against visually similar majority class (Healthy vs Mild NPDR) - Default batch size reduced 64->32 to accommodate larger images - 4-epoch warmup-only test: macro F1 0.45, weighted F1 0.74, kappa 0.62 already exceeding previous 22-epoch best on all metrics 9dd8b3007e79f0634f868d405f5e7c48917305fd 2026-04-20T14:03:17Z 2026-04-20T14:03:17Z Vineet Kumar git@vineetk.net commit 9dd8b3007e79f0634f868d405f5e7c48917305fd parent 8f571ee51854540d186aafc01d25908bba5ea09e Author: Vineet Kumar <git@vineetk.net> Date: Mon, 20 Apr 2026 10:03:17 -0400 improve training for class-imbalanced DR classification - WeightedRandomSampler for balanced class sampling during training - FocalLoss (gamma=2) for hard-example mining, no alpha (sampler handles balance) - Stronger augmentation: RandomResizedCrop, ColorJitter, GaussianBlur, 30deg rotation - Discriminative LR: backbone at 0.1x, head at 1x - 3-epoch linear warmup + cosine decay scheduler - Early stopping on macro F1 instead of weighted F1 - Updated defaults: batch_size=64, epochs=50, patience=10 - Comparison chart now shows both macro and weighted F1 - DataParallel checkpoint handling (strip module. prefix) - Enable cudnn.benchmark for training speed 8f571ee51854540d186aafc01d25908bba5ea09e 2026-04-19T01:34:11Z 2026-04-19T01:34:11Z Vineet Kumar git@vineetk.net commit 8f571ee51854540d186aafc01d25908bba5ea09e Author: Vineet Kumar <git@vineetk.net> Date: Sat, 18 Apr 2026 21:34:11 -0400 initial commit