Closed hannes56a closed 3 years ago
Nobody has an idea??
When calling the detect_image.py script, did you try setting --threshold=0?
Hi, i will check it, but if this is only the threshold for detection output, I tried with 0.2 and the "not detected" objects have a score above 0.5 ...
I tried with --threshold=0, but it doesn´t help...
@hannes56a can post your full pipeline.conf?
[pipeline.config.txt](https://github.com/google-coral/tflite/files/6149571/pipeline.config.txt)
# Copyright 2020 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
model {
ssd {
num_classes: 4
image_resizer {
fixed_shape_resizer {
height: 300
width: 300
}
}
feature_extractor {
type: "ssd_mobilenet_v1"
depth_multiplier: 1.0
min_depth: 16
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.00999999977648
}
}
activation: RELU_6
batch_norm {
decay: 0.97000002861
center: true
scale: true
epsilon: 0.0010000000475
}
}
override_base_feature_extractor_hyperparams: true
}
box_coder {
faster_rcnn_box_coder {
y_scale: 10.0
x_scale: 10.0
height_scale: 5.0
width_scale: 5.0
}
}
matcher {
argmax_matcher {
matched_threshold: 0.5
unmatched_threshold: 0.5
ignore_thresholds: false
negatives_lower_than_unmatched: true
force_match_for_each_row: true
use_matmul_gather: true
}
}
similarity_calculator {
iou_similarity {
}
}
box_predictor {
convolutional_box_predictor {
conv_hyperparams {
regularizer {
l2_regularizer {
weight: 3.99999989895e-05
}
}
initializer {
random_normal_initializer {
mean: 0.0
stddev: 0.00999999977648
}
}
activation: RELU_6
batch_norm {
decay: 0.97000002861
center: true
scale: true
epsilon: 0.0010000000475
}
}
min_depth: 0
max_depth: 0
num_layers_before_predictor: 0
use_dropout: false
dropout_keep_probability: 0.800000011921
kernel_size: 1
box_code_size: 4
apply_sigmoid_to_scores: false
class_prediction_bias_init: -4.59999990463
}
}
anchor_generator {
ssd_anchor_generator {
num_layers: 6
min_scale: 0.20000000298
max_scale: 0.949999988079
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
aspect_ratios: 3.0
aspect_ratios: 0.333299994469
}
}
post_processing {
batch_non_max_suppression {
score_threshold: 0.300000011921
iou_threshold: 0.600000023842
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
normalize_loss_by_num_matches: true
loss {
localization_loss {
weighted_smooth_l1 {
}
}
classification_loss {
weighted_sigmoid_focal {
gamma: 2.0
alpha: 0.75
}
}
classification_weight: 1.0
localization_weight: 1.0
}
encode_background_as_zeros: true
normalize_loc_loss_by_codesize: true
inplace_batchnorm_update: true
freeze_batchnorm: false
}
}
train_config {
batch_size: 64
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
random_vertical_flip {
}
}
data_augmentation_options {
ssd_random_crop {
}
}
data_augmentation_options {
random_rotation90 {
}
}
data_augmentation_options {
random_adjust_contrast {
}
}
#data_augmentation_options {
# rgb_to_gray {
# }
#}
data_augmentation_options {
random_image_scale {
}
}
data_augmentation_options {
random_adjust_brightness {
}
}
data_augmentation_options {
random_adjust_hue {
}
}
data_augmentation_options {
random_adjust_saturation {
}
}
sync_replicas: true
optimizer {
momentum_optimizer {
learning_rate {
cosine_decay_learning_rate {
learning_rate_base: 0.40000000298
total_steps: 60000
warmup_learning_rate: 0.0599999986589
warmup_steps: 500
}
}
momentum_optimizer_value: 0.899999976158
}
use_moving_average: false
}
fine_tune_checkpoint: "/tensorflow/models/research/learn_pet/ckpt/model.ckpt"
from_detection_checkpoint: true
load_all_detection_checkpoint_vars: true
num_steps: 50000
startup_delay_steps: 0.0
replicas_to_aggregate: 8
max_number_of_boxes: 100
unpad_groundtruth_tensors: false
}
train_input_reader {
label_map_path: "/tensorflow/models/research/learn_pet/muster_cobot/muster_cobot_label_map.pbtxt"
tf_record_input_reader {
input_path: "/tensorflow/models/research/learn_pet/muster_cobot/train.tfrecords"
}
}
eval_config {
num_examples: 8000
metrics_set: "coco_detection_metrics"
use_moving_averages: false
num_visualizations: 15
}
eval_input_reader {
label_map_path: "/tensorflow/models/research/learn_pet/muster_cobot/muster_cobot_label_map.pbtxt"
shuffle: false
num_readers: 1
tf_record_input_reader {
input_path: "/tensorflow/models/research/learn_pet/muster_cobot/eval.tfrecords"
}
}
graph_rewriter {
quantization {
delay: 48000
weight_bits: 8
activation_bits: 8
}
}
Actually, that pipeline looks good, on the step where you run the export_tflite_ssd_graph.py
make sure you also add this flag:
--max_detections=100
since for what ever odd reason, tensorflow decided that it'd default to 10: https://github.com/tensorflow/models/blob/master/research/object_detection/export_tflite_ssd_graph.py#L107
I guess for you, since you are following the tutorial, that means you need to modify this scipt: https://github.com/google-coral/tutorials/blob/master/docker/object_detection/scripts/convert_checkpoint_to_edgetpu_tflite.sh#L55
Thanks a lot! Looks like this is the reason for sure. I cant test it now but I will report, when i have test it! Again: Thanks a lot!
It works! Thanks a lot again!
Hi all, I retrained the ssd_mobilNet_v1 model with own data (4 classes) (following the google tutorial for the coral chip: https://coral.ai/docs/edgetpu/retrain-detection/#using-the-coral-usb-accelerator).
Now I´m testing the new model on my test images with the "detect_image.py" script. It basically works, but I got maximum 10 detections (boundingboxes). There are many more objects to find. I tested some things and there are every time maximum 10 detections in all (not 10 detections per class). If i "overpaint" these detections in the testimage, the model finds another 10 detections...
Perhaps some usefull information:
"max_detections_per_class" and "max_total_detections" in the "pipeline.config" for the retraining are both 100.
I do not find any parameter or reason for this behavior. Can please someone help me?
Greeting, Hannes