do not reproduce the yolox-nano result

[Tim5Tang]

use the cmd and parameter as follow:

python tools/train.py -n yolox-nano -d 2 -b 64 --fp16 -o --cache

2021-09-08 14:42:06.344 | INFO | yolox.core.trainer:before_train:126 - args: Namespace(batch_size=64, cache=True, ckpt=None, devices=1, dist_backend='nccl', dist_url=None, exp_file=None, experiment_name='nano', fp16=True, machine_rank=0, name='yolox-nano', num_machines=1, occupy=True, opts=[], resume=False, start_epoch=None)
2021-09-08 14:42:06.347 | INFO | yolox.core.trainer:before_train:127 - exp value:
╒══════════════════╤════════════════════════════╕
│ keys │ values │
╞══════════════════╪════════════════════════════╡
│ seed │ None │
├──────────────────┼────────────────────────────┤
│ output_dir │ './YOLOX_outputs' │
├──────────────────┼────────────────────────────┤
│ print_interval │ 10 │
├──────────────────┼────────────────────────────┤
│ eval_interval │ 10 │
├──────────────────┼────────────────────────────┤
│ num_classes │ 80 │
├──────────────────┼────────────────────────────┤
│ depth │ 0.33 │
├──────────────────┼────────────────────────────┤
│ width │ 0.25 │
├──────────────────┼────────────────────────────┤
│ data_num_workers │ 4 │
├──────────────────┼────────────────────────────┤
│ input_size │ (416, 416) │
├──────────────────┼────────────────────────────┤
│ multiscale_range │ 5 │
├──────────────────┼────────────────────────────┤
│ data_dir │ None │
├──────────────────┼────────────────────────────┤
│ train_ann │ 'instances_train2017.json' │
├──────────────────┼────────────────────────────┤
│ val_ann │ 'instances_val2017.json' │
├──────────────────┼────────────────────────────┤
│ mosaic_prob │ 0.5 │
├──────────────────┼────────────────────────────┤
│ mixup_prob │ 1.0 │
├──────────────────┼────────────────────────────┤
│ hsv_prob │ 1.0 │
├──────────────────┼────────────────────────────┤
│ flip_prob │ 0.5 │
├──────────────────┼────────────────────────────┤
│ degrees │ 10.0 │
├──────────────────┼────────────────────────────┤
│ translate │ 0.1 │
├──────────────────┼────────────────────────────┤
│ mosaic_scale │ (0.5, 1.5) │
├──────────────────┼────────────────────────────┤
│ mixup_scale │ (0.5, 1.5) │
├──────────────────┼────────────────────────────┤
│ shear │ 2.0 │
├──────────────────┼────────────────────────────┤
│ perspective │ 0.0 │
├──────────────────┼────────────────────────────┤
│ enable_mixup │ False │
├──────────────────┼────────────────────────────┤
│ warmup_epochs │ 5 │
├──────────────────┼────────────────────────────┤
│ max_epoch │ 300 │
├──────────────────┼────────────────────────────┤
│ warmup_lr │ 0 │
├──────────────────┼────────────────────────────┤
│ basic_lr_per_img │ 0.00015625 │
├──────────────────┼────────────────────────────┤
│ scheduler │ 'yoloxwarmcos' │
├──────────────────┼────────────────────────────┤
│ no_aug_epochs │ 15 │
├──────────────────┼────────────────────────────┤
│ min_lr_ratio │ 0.05 │
├──────────────────┼────────────────────────────┤
│ ema │ True │
├──────────────────┼────────────────────────────┤
│ weight_decay │ 0.0005 │
├──────────────────┼────────────────────────────┤
│ momentum │ 0.9 │
├──────────────────┼────────────────────────────┤
│ exp_name │ 'nano' │
├──────────────────┼────────────────────────────┤
│ test_size │ (416, 416) │
├──────────────────┼────────────────────────────┤
│ test_conf │ 0.01 │
├──────────────────┼────────────────────────────┤
│ nmsthre │ 0.65 │
├──────────────────┼────────────────────────────┤
│ random_size │ (10, 20) │
╘══════════════════╧════════════════════════════╛
2021-09-08 14:42:06.468 | INFO | yolox.core.trainer:before_train:133 - Model Summary: Params: 0.91M, Gflops: 1.08
2021-09-08 14:42:08.182 | INFO | yolox.data.datasets.coco:init:45 - loading annotations into memory...
2021-09-08 14:42:20.787 | INFO | yolox.data.datasets.coco:init:45 - Done (t=12.60s)
2021-09-08 14:42:20.787 | INFO | pycocotools.coco:init:92 - creating index...
2021-09-08 14:42:21.736 | INFO | pycocotools.coco:init:92 - index created!
2021-09-08 14:42:50.371 | WARNING | yolox.data.datasets.coco:_cache_images:69 -

but the result is :

2021-09-10 05:46:06.608 | INFO | yolox.core.trainer:save_ckpt:318 - Save weights to ./YOLOX_outputs/nano
2021-09-10 05:46:16.802 | INFO | yolox.evaluators.coco_evaluator:evaluate_prediction:171 - Evaluate in main process...
2021-09-10 05:46:20.158 | INFO | yolox.evaluators.coco_evaluator:evaluate_prediction:204 - Loading and preparing results...
2021-09-10 05:46:24.156 | INFO | yolox.evaluators.coco_evaluator:evaluate_prediction:204 - DONE (t=4.00s)
2021-09-10 05:46:24.156 | INFO | pycocotools.coco:loadRes:433 - creating index...
2021-09-10 05:46:24.287 | INFO | pycocotools.coco:loadRes:433 - index created!
2021-09-10 05:46:36.375 | INFO | yolox.core.trainer:evaluate_and_save_model:309 -
Average forward time: 0.52 ms, Average NMS time: 0.56 ms, Average inference time: 1.08 ms
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.238
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.388
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.246
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.074
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.249
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.383
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.228
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.351
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.375
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.128
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.407
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.579

2021-09-10 05:46:36.376 | INFO | yolox.core.trainer:save_ckpt:318 - Save weights to ./YOLOX_outputs/nano
2021-09-10 05:46:36.418 | INFO | yolox.core.trainer:after_train:184 - Training of experiment is done and the best AP is 23.83

[Joker316701882]

@Tim5Tang With this version, we can get the best mAP around 24.

However, the results reported in our tech report and this repo are based on an ImageNet pre-trained model. That's why you can't reproduce the result. But noted that yolox_nano is the only one that adopts the pretrain model.

[Tim5Tang]

ok, i see thanks.