Feasibility to evaluate DeepLab-MobileNetV3 model(s)

[vbhvdnsh]

I am trying to run evaluation on DeepLab based on Mobilenet-V3 for which I am accessing the checkpoint corresponding to mobilenetv3_small_cityscapes_trainfine. The code used for the same is obtained from the same repository; however I am unable to successfully run the evaluation as the system runs into errors at different stages of the process. It seems that most of these are concerning the Tensorflow and Python versioning, and while I have downgraded to Tensorflow version 1.14 with Python version 3.6.13 but code eventually crashes at: summary_op = tf.compat.v1.summary.merge(summary_ops)
Barring the input "samples" variable, I am unable to observe the actual data-values in other variables of interest, like 'predictions', 'metric_map' as despite running global_variables_initializer() these complain about uninitialized variables issues. It is also not clear where/when the system accesses the checkpoint path specified through one of the FLAG properties.
To summarize, I am unsure, based on this being quite an old model, if the code base is still functional for executing this model using the provided checkpoint (*.pb)!!