MindCV 0.2.2
What's Changed
- feat: add model script, training recipe and trained weights of ConvNeXt V2 by @The-truthh in #654
- feat: support gradient accumulation function by @zhtmike in #609
- refactor: modify filenames of pvtv2, inceptionv3, inceptionv4 and shufflenetv1&v2 by @XixinYang in #677
- feat: support DynamicLossScale for TrainStep by @geniuspatrick in #678
- feat: coat_mini update by @SunJuil-Ty in #662
- feat: add OneCycleLR and CyclicLR scheduler by @The-truthh in #657
- feat: better logging by @geniuspatrick in #684
- feat: add training recipe and pretrained weights of mnasnet_1.3 by @XixinYang in #679
- feat: support multi-scale feature extraction. by @DexterJZ in #643
- hotfix: fix the train_step optimizer.global_step type (int32 -> float32) bug by @The-truthh in #686
- feat: upgrade ms to v2.0 while maintaining compatibility of v1.8 by @geniuspatrick in #674
- fix: fix the bug of mobilenet_v3 on ms1.8 on Ascend in deployment tutorial by @The-truthh in #688
- feat: update training recipes and pretrained weights of shufflenet_v2 (amp_level: O3) by @The-truthh in #681
- fix: correct variable "keep_prob" to "drop_path_prob" of efficientnet by @wcrzlh in #672
- hotfix: top_k save policy; coat_mini & coat_tiny recipe by @geniuspatrick in #689
- docs: update links in markdown files by @XixinYang in #687
- bump version to 0.2.2 by @geniuspatrick in #691
New Contributors
Full Changelog: v0.2.1...v0.2.2
Contributors
zhtmike, DexterJZ, and 5 other contributors