haritsahm
commented
2 years ago I have tested it with partial quantization and another issue also occured.
Partial Quantization Export
Accumulating evaluation results...
DONE (t=1.51s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.574
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.810
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.616
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.275
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.576
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.772
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.199
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.587
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.691
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.452
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.729
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.847
Skip Layer detect.proj_conv
op amax = 5.1035, amax = -1.0000
op amax = 5.1425, amax = 5.1035
amax = 5.1425
op amax = 6.0546, amax = -1.0000
op amax = 3.9350, amax = 6.0546
Not quantable op, skip
op amax = 3.9134, amax = -1.0000
op amax = 3.9041, amax = 3.9134
Not quantable op, skip
op amax = 11.4002, amax = -1.0000
op amax = 12.6850, amax = 11.4002
amax = 12.6850
op amax = 8.2758, amax = -1.0000
op amax = 9.5599, amax = 8.2758
amax = 9.5599
op amax = 3.9648, amax = -1.0000
op amax = 3.9648, amax = 3.9648
amax = 3.9648
op amax = 4.4959, amax = -1.0000
op amax = 4.4959, amax = 4.4959
amax = 4.4959
op amax = 3.9817, amax = -1.0000
op amax = 3.9817, amax = 3.9817
amax = 3.9817
Inferencing model in val datasets.: 100%
loading annotations into memory...
Done (t=0.03s)
creating index...
index created!
Loading and preparing results...
DONE (t=1.13s)
creating index...
index created!
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=14.29s).
Accumulating evaluation results...
DONE (t=1.59s).
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.574
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.810
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.617
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.275
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.577
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.773
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.199
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.588
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.692
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.455
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.730
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.847
(0.8100440948596181, 0.5744896506105379)The exported model will cause an error in TensorRT
[10/20/2022-15:01:22] [V] [TRT] QuantizeLinear_25 [QuantizeLinear] inputs: [backbone.ERBlock_2.0.conv_1x1.weight -> (64, 32, 1, 1)[FLOAT]], [611 -> (64)[FLOAT]], [2028 -> (64)[INT8]],
[10/20/2022-15:01:22] [V] [TRT] Registering layer: backbone.ERBlock_2.0.conv_1x1.weight for ONNX node: backbone.ERBlock_2.0.conv_1x1.weight
[10/20/2022-15:01:22] [E] [TRT] parsers/onnx/ModelImporter.cpp:791: While parsing node number 25 [QuantizeLinear -> "614"]:
[10/20/2022-15:01:22] [E] [TRT] parsers/onnx/ModelImporter.cpp:792: --- Begin node ---
[10/20/2022-15:01:22] [E] [TRT] parsers/onnx/ModelImporter.cpp:793: input: "backbone.ERBlock_2.0.conv_1x1.weight"
input: "611"
input: "2028"
output: "614"
name: "QuantizeLinear_25"
op_type: "QuantizeLinear"
attribute {
name: "axis"
i: 0
type: INT
}
[10/20/2022-15:01:22] [E] [TRT] parsers/onnx/ModelImporter.cpp:794: --- End node ---
[10/20/2022-15:01:22] [E] [TRT] parsers/onnx/ModelImporter.cpp:796: ERROR: parsers/onnx/builtin_op_importers.cpp:1150 In function QuantDequantLinearHelper:
[6] Assertion failed: scaleAllPositive && "Scale coefficients must all be positive"
[10/20/2022-15:01:22] [E] Failed to parse onnx fileTensorRT Commands
trtexec --onnx=best_ckpt_partial_dynamic.onnx --saveEngine=best_ckpt_partial_dynamic.engine --minShapes=images:1x3x640x640 --optShapes=images:4x3x640x640 --maxShapes=images:8x3x640x640 --fp16 --int8 --warmUp=1000 --avgRuns=1000 --workspace=2048 --inputIOFormats=fp16:chw --verbose
lippman1125
Before Asking
[X] I have read the README carefully. 我已经仔细阅读了README上的操作指引。
[X] I want to train my custom dataset, and I have read the tutorials for training your custom data carefully and organize my dataset correctly; (FYI: We recommand you to apply the config files of xx_finetune.py.) 我想训练自定义数据集，我已经仔细阅读了训练自定义数据的教程，以及按照正确的目录结构存放数据集。（FYI: 我们推荐使用xx_finetune.py等配置文件训练自定义数据集。）
[X] I have pulled the latest code of main branch to run again and the problem still existed. 我已经拉取了主分支上最新的代码，重新运行之后，问题仍不能解决。
Search before asking
Question
After training the model following the guide in tutorial_repopt, I got an unusual behaviour when running the PTQ and QAT training steps, similar to what I reported in 535#issuecomment-1284808881.
Notes:
Final numbers of valid images: 64094/ labels: 64094.Final numbers of valid images: 2693/ labels: 2693.preds, s_featmaps = self.model(images)are tensor ofnansCommands
Configs
Eval Output
The PTQ Output:
Additional
No response