Jialiang Ma

Hi~， I notice that local feature (lf) variable is different in training phase and test phase. Specifically, in "AlignedReID/models/ResNet.py", ``` if not self.training: lf = self.horizon_pool(x) if self.aligned and self.training:...

The actual performance is far from the report

3

Hi, I trained Market1501 with resnet50 AlignedReID, but the final result is only 87.9% rank-1 / 72.0% mAP (without re-ranking), 89.3% rank-1 / 85.4% mAP (with re-ranking), which is **far...

为什么在layers/__init__.py的make_loss_with_center函数中给center loss乘了cfg.SOLVER.CENTER_LOSS_WEIGHT，还要在engine/trainer中的create_supervised_trainer_with_center处给center_criterion的param.grad乘个（1. / cfg.SOLVER.CENTER_LOSS_WEIGHT）？这样cfg.SOLVER.CENTER_LOSS_WEIGHT不就没有作用了吗？

Hi, I can't reproduce your work on the market1501 dataset. Specifically, I use resnet50 as the backbone and without rerank, I can only get results of mAP 84.1% / rank-1...

Pretrained Weights.

1

Thanks for sharing the code base, can you please share the pretrained model or trained weight file? Thanks in advance.

I have a question, if I use the classifier trained by LSUN cat, can it be used together with the diffusion model trained by LSUN bedroom? How's the performance?

In the run scripts, 256x256 model (unconditional) as follows: `MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16...