Commit Graph

  • 2410264a00 dd kwargs for naflexvit, needs revisit for nn.Parameters Ross Wightman 2025-09-29 14:04:17 -07:00
  • 1e65174a03 Fix DarkStage device kwargs Ross Wightman 2025-09-29 13:54:59 -07:00
  • 6e01090baa avg pool should not have been passed dd Ross Wightman 2025-09-29 11:52:43 -07:00
  • 63d4dc9a1c Add device/dtype factory kwargs to beit, efficientformer*, efficientvit*, focalnet, levit. Fix ByobNet Ross Wightman 2025-09-29 10:53:36 -07:00
  • 5969d0891d Add dd factory kwargs to byobnet, cspnet, davit, edgenext Ross Wightman 2025-09-29 07:21:42 -07:00
  • f9afb84017 Fix typo for s2d norm Ross Wightman 2025-09-28 16:17:39 -07:00
  • 9dea868f51 Add dd factory kwargs to all EfficientNetBuilder models, MobileNet V1/V2/V3/V4/V5, EfficientNet, etc Ross Wightman 2025-09-28 15:39:20 -07:00
  • 481a693a59 dd factory kwargs for fastvit, convnext, mambaout Ross Wightman 2025-09-28 15:04:15 -07:00
  • 8721e1a04a Add dd factory kwargs to nfnet and resnetv2 Ross Wightman 2025-09-28 11:20:00 -07:00
  • 648e21826a Add dd factory kwargs to maxxvit and regnet Ross Wightman 2025-09-27 15:35:58 -07:00
  • f5c1cbdbf2 Add dd to other ResNet based models, Res2Net, ResNeSt, SKNet Ross Wightman 2025-09-27 09:53:59 -07:00
  • a09e4d21cc Add dd factory kwargs to eva, resnet Ross Wightman 2025-09-27 08:10:45 -07:00
  • ad57dce01a Adding dd factory_kwargs to modules in timm/layers, initial model WIP in vision_transformer.py Ross Wightman 2025-09-26 13:39:47 -07:00
  • d3fdea8a1d Typing, super(), buffer dtype fixes for timm/layers and timm/models dd_factory_kwargs Ross Wightman 2025-10-02 07:36:17 -07:00
  • 5cadf13399 More dd arg conversions. fasternet, gcvit, hgnet, nextvit, starnet, vision_transformer_hybrid/relpos/sam, vitamin. Temp fix for shvit reset_classifier (need better approach) Ross Wightman 2025-10-01 16:18:56 -07:00
  • 21b1ae7e29 More dd factory kwargs updates. hiera, hieradet_sam2, metaformer, mlp_mixer, mobilevit, pnasnet, rexnet, sequencer, shvit model files. Fixed blur_pool dtype/device handling and update Mlp modules w/ annotations and a fix. Ross Wightman 2025-10-01 14:07:35 -07:00
  • 53caeb0394 Add some more dd kwarg updates, crossvit, ghostnet, rdnet, repghost, repvit, selecsls, swiftformer Ross Wightman 2025-10-01 12:43:56 -07:00
  • c7955eb86b Add dd factory kwargs to all swin transformers and volo Ross Wightman 2025-09-30 14:35:24 -07:00
  • 6a3342ca8a dd factory kwargs added to a bunch of vit/vit-hybrids. cait, coat, convit, convmixer, deit, mvitv2, nest, pit, pvt_v2, tiny_vit, tnt, twins, visformer, xcit Ross Wightman 2025-09-30 10:01:07 -07:00
  • 0645384b3a Review huggingface_hub integration Wauplin 2025-09-30 15:05:20 +02:00
  • 068e6d4411 Remove **dd from two inception reset_classifier calls Ross Wightman 2025-09-29 16:02:09 -07:00
  • a7dc50fd1a A whack of classic convnets converted with dd factory kwargs. densenet, dla, dpn, hrnet, inception_next, inception_resnet_v2, inception_v3/v4, senet, tresnet, vgg, vovnet, xception, xception_aligned Ross Wightman 2025-09-29 14:04:45 -07:00
  • 1e172a0b1f dd kwargs for naflexvit, needs revisit for nn.Parameters Ross Wightman 2025-09-29 14:04:17 -07:00
  • 8cbbf396a3 Fix DarkStage device kwargs Ross Wightman 2025-09-29 13:54:59 -07:00
  • 3a85ed4b76 avg pool should not have been passed dd Ross Wightman 2025-09-29 11:52:43 -07:00
  • 4c35b78d9c Add device/dtype factory kwargs to beit, efficientformer*, efficientvit*, focalnet, levit. Fix ByobNet Ross Wightman 2025-09-29 10:53:36 -07:00
  • f15f7c9104 Add dd factory kwargs to byobnet, cspnet, davit, edgenext Ross Wightman 2025-09-29 07:21:42 -07:00
  • 4d19b341fc Fix typo for s2d norm Ross Wightman 2025-09-28 16:17:39 -07:00
  • 60db539bc6 Add dd factory kwargs to all EfficientNetBuilder models, MobileNet V1/V2/V3/V4/V5, EfficientNet, etc Ross Wightman 2025-09-28 15:39:20 -07:00
  • 10e7020165 dd factory kwargs for fastvit, convnext, mambaout Ross Wightman 2025-09-28 15:04:15 -07:00
  • ee751efe96 Add dd factory kwargs to nfnet and resnetv2 Ross Wightman 2025-09-28 11:20:00 -07:00
  • b94c221862 Add dd factory kwargs to maxxvit and regnet Ross Wightman 2025-09-27 15:35:58 -07:00
  • 325a6cc5f8 Add dd to other ResNet based models, Res2Net, ResNeSt, SKNet Ross Wightman 2025-09-27 09:53:59 -07:00
  • 90a35c8853 Add dd factory kwargs to eva, resnet Ross Wightman 2025-09-27 08:10:45 -07:00
  • 7560ca5962 Adding dd factory_kwargs to modules in timm/layers, initial model WIP in vision_transformer.py Ross Wightman 2025-09-26 13:39:47 -07:00
  • e7bd97be7e Add missing pretrained_cfg for old untrained model Ross Wightman 2025-09-24 07:21:38 -07:00
  • 17af914b50 Fix a bug for unused x/h effnets Ross Wightman 2025-09-23 22:32:36 -07:00
  • 3f58765abf Add calculate_drop_path_rates helper. Make force calcs on cpu device to avoid issue with device contexts/meta device use. Fix a few small inconsistencies that were noticed. Ross Wightman 2025-09-23 21:20:26 -07:00
  • 008c12f149 Add missing pretrained_cfg for old untrained model calculate_dpr Ross Wightman 2025-09-24 07:21:38 -07:00
  • 2c8c3382fb Fix a bug for unused x/h effnets Ross Wightman 2025-09-23 22:32:36 -07:00
  • 31ca21637f Add calculate_drop_path_rates helper. Make force calcs on cpu device to avoid issue with device contexts/meta device use. Fix a few small inconsistencies that were noticed. Ross Wightman 2025-09-23 21:20:26 -07:00
  • a510490280 Dev version 1.0.21.dev0 Ross Wightman 2025-09-23 21:19:42 -07:00
  • 30b757ccb0 Update CONTRIBUTING.md Ross Wightman 2025-09-22 22:24:45 -07:00
  • 234907eca5 Release 1.0.20 v1.0.20 Ross Wightman 2025-09-21 10:25:15 -07:00
  • c6837bd1b9 Correct lvd_1689m / sat_493m -> lvd1689m / sat493m for consistency with other models, matches hub rename Ross Wightman 2025-09-21 10:00:25 -07:00
  • 05152943e5 Fix #2577, resize keep ratio repr Ross Wightman 2025-09-20 17:13:15 -07:00
  • 019550eeaf Remove unavailable weight Ross Wightman 2025-09-18 09:20:00 -07:00
  • 5c8c5dfcdd Update README.md Ross Wightman 2025-09-17 14:47:06 -07:00
  • f55ca98d4e Fix builder variant for dinov3 qkvb h/16 Ross Wightman 2025-09-17 13:10:48 -07:00
  • e423371922 Pass task arg throught to hf hub push Ross Wightman 2025-09-17 11:56:58 -07:00
  • 48bf66b27c Final dinov3 pretrained cfgs, pointing to uploaded weights Ross Wightman 2025-09-17 11:56:18 -07:00
  • 33bd6ad9b6 Move make_coords in dinov3 RoPE to a free fn so it can be wrapped for fx Ross Wightman 2025-09-16 16:29:01 -07:00
  • 112b2cbe8c Remove '224' from dinov3 model names and enable dynamic_img_size=True by default. Add assert in DINOv3 ROPE module to avoid torchscript failure (hopefully no more?) Ross Wightman 2025-09-16 14:20:34 -07:00
  • 0093117fc4 Missed 7b filter from GITHUB specific Ross Wightman 2025-09-16 11:38:16 -07:00
  • 8104d3cd32 Further 7b test filtering. Remove ref_feat_shape from DINOv3 RoPE as it's normalized. Fix torchscript issue with + unary op Ross Wightman 2025-09-16 10:57:05 -07:00
  • d5547b4ff2 Exclude 7b from unit tests Ross Wightman 2025-09-15 17:13:30 -07:00
  • 18cb3bc16f Add DINOv3 model defs, add EVA additions to support DINOv3. Currently testing... Ross Wightman 2025-09-15 16:38:05 -07:00
  • 1a49c564a2 Remove hard-coded min/max period values that were used for testing, clarify comment Ross Wightman 2025-09-15 12:32:10 -07:00
  • 91e8dc56e8 DINOv3 rotary position embedding impl Ross Wightman 2025-09-15 11:29:53 -07:00
  • 5d5707adde Add DINOv3 ConvNeXt weights. Ross Wightman 2025-09-11 11:17:36 -07:00
  • 9a44a236c1 Fix builder variant for dinov3 qkvb h/16 dinov3 Ross Wightman 2025-09-17 13:10:48 -07:00
  • 04851a0548 Pass task arg throught to hf hub push Ross Wightman 2025-09-17 11:56:58 -07:00
  • 9bd8137f80 Final dinov3 pretrained cfgs, pointing to uploaded weights Ross Wightman 2025-09-17 11:56:18 -07:00
  • 7fbebaabd1 Move make_coords in dinov3 RoPE to a free fn so it can be wrapped for fx Ross Wightman 2025-09-16 16:29:01 -07:00
  • 508df64fd9 Remove '224' from dinov3 model names and enable dynamic_img_size=True by default. Add assert in DINOv3 ROPE module to avoid torchscript failure (hopefully no more?) Ross Wightman 2025-09-16 14:20:34 -07:00
  • 2b1e26637e Missed 7b filter from GITHUB specific Ross Wightman 2025-09-16 11:38:16 -07:00
  • 28278f14c4 Further 7b test filtering. Remove ref_feat_shape from DINOv3 RoPE as it's normalized. Fix torchscript issue with + unary op Ross Wightman 2025-09-16 10:57:05 -07:00
  • f133fb3f5c Exclude 7b from unit tests Ross Wightman 2025-09-15 17:13:30 -07:00
  • e0592d4669 Add DINOv3 model defs, add EVA additions to support DINOv3. Currently testing... Ross Wightman 2025-09-15 16:38:05 -07:00
  • a9018fffbc Remove hard-coded min/max period values that were used for testing, clarify comment Ross Wightman 2025-09-15 12:32:10 -07:00
  • fe843da68e DINOv3 rotary position embedding impl Ross Wightman 2025-09-15 11:29:53 -07:00
  • a1794ccc18 Add DINOv3 ConvNeXt weights. Ross Wightman 2025-09-11 11:17:36 -07:00
  • 2d1bc8a8c3 First conv is different for fastvit mci3/mci4 models Ross Wightman 2025-09-10 17:09:12 -07:00
  • 639d6ab811 mci3/mci4 backbones are stride 64, so 4x4 pooler output Ross Wightman 2025-09-10 15:55:56 -07:00
  • b8cc442b64 Add MobileCLIP2 image encoders Ross Wightman 2025-09-10 14:29:03 -07:00
  • 5e9d4b374b First conv is different for fastvit mci3/mci4 models apple_mobileclip2 Ross Wightman 2025-09-10 17:09:12 -07:00
  • f251042609 mci3/mci4 backbones are stride 64, so 4x4 pooler output Ross Wightman 2025-09-10 15:55:56 -07:00
  • f242d23a15 Add MobileCLIP2 image encoders Ross Wightman 2025-09-10 14:29:03 -07:00
  • 818663b8b6 Small tweak, no change in execution if addition features not used. Ross Wightman 2025-09-04 12:25:18 -07:00
  • 63277546f7 Allow user to specify additional features to be returned by Image Dataset when using ReaderHfds Augustin Godinot 2025-08-18 23:19:55 -04:00
  • cedba69c19 Merge branch 'ha405-enhanced_validate.py' Ross Wightman 2025-09-04 08:48:37 -07:00
  • 40042b7d9c f1/p/r -> 0..100 to match existing accuracy range Ross Wightman 2025-09-04 08:47:38 -07:00
  • 54b359376b Merge branch 'enhanced_validate.py' of github.com:ha405/pytorch-image-models into ha405-enhanced_validate.py Ross Wightman 2025-09-04 08:24:47 -07:00
  • 6d30b82d3f fix: create results_dir if missing before saving results zhima 2025-08-27 16:19:20 +08:00
  • 74425f99d8 feat(validate): add extra metrics, JSON export, and latency/memory tracking DASH 2025-08-15 07:40:54 +05:00
  • 954613a470 Never passed torchcompile-mode arg through to compile call Ross Wightman 2025-08-08 12:22:57 -07:00
  • 055a600ed9 Fix #2562, add siglip2 naflex vit encoder weights Ross Wightman 2025-08-04 11:57:59 -07:00
  • 0719dc3df3 Fix #2562, add siglip2 naflex vit encoder weights naflex_siglip2_weights Ross Wightman 2025-08-04 11:57:59 -07:00
  • b3869b6283 Update comments for mnv5 pretrained configs Ross Wightman 2025-08-01 21:06:29 -07:00
  • 6c5e3ce3d0 Add support for Gemma 3n MobileNetV5 encoder weight loading Ross Wightman 2025-08-01 13:58:59 -07:00
  • dddb914ab9 Update comments for mnv5 pretrained configs gemma3n_enc_weights Ross Wightman 2025-08-01 21:06:29 -07:00
  • 382baf9713 Add support for Gemma 3n MobileNetV5 encoder weight loading Ross Wightman 2025-08-01 13:58:59 -07:00
  • 1e1c637aaf Dev version 1.0.20.dev0 Ross Wightman 2025-08-01 12:42:46 -07:00
  • 2ddbe5ed2a Add mobileclip2 encoder weights Ross Wightman 2025-08-01 09:33:53 -07:00
  • 435ab7a882 Pass init_values at hieradet_sam2 Ofer Hasson 2025-07-26 06:51:27 +03:00
  • 55d4b5ddf9 Train script issues. Fix patch_size extraction in naflex mode + distributed. Fix update time calculations after adding naflex mode. Fix #2556 Ross Wightman 2025-07-24 13:19:17 -07:00
  • d08d5a0392 Fix #2555 for Python 3.9 compat. Release 1.0.19 v1.0.19 Ross Wightman 2025-07-23 20:02:29 -07:00
  • 0e79579323 Fix newer type annotation in layers/typing Ross Wightman 2025-07-23 15:21:07 -07:00
  • e6ab6bc3c6 Release 1.0.18 v1.0.18 Ross Wightman 2025-07-23 12:33:44 -07:00
  • 280a90c661 Support set_input_size() in EVA models (#2554) Ross Wightman 2025-07-22 21:42:27 -07:00